2025-11-03 19:59:18 +08:00
""" Blob storage connector """
2026-07-03 12:53:39 +08:00
2025-11-03 19:59:18 +08:00
import logging
import os
feat(connectors): ETag-based bypass for incremental S3 ingestion (#14628) (#14677)
### What problem does this PR solve?
S3-family connector syncs currently re-download every in-window object
just so we can compute `xxhash128(blob)` and compare against
`Document.content_hash`. Anything that bumps `LastModified` without
changing bytes (`aws s3 cp` touches, bucket re-encryption, etc.) pays
full bandwidth and re-parses files that didn't actually change. #14628
covers the broader incremental-ingestion redesign; this PR is the first
slice.
The fix is a pre-listing short-circuit. `BlobStorageConnector` (S3 / R2
/ GCS / OCI / S3-compat) now implements a new `FingerprintConnector`
interface: `list_keys()` paginates `list_objects_v2` and yields
`KeyRecord(key, fingerprint)` where `fingerprint = xxhash128(ETag)`. The
orchestrator joins those against the connector's existing `{doc_id:
content_hash}` map and only calls `get_value(key)` when the fingerprint
differs. Unchanged keys are skipped entirely — no `GetObject`, no
re-parse.
No DDL. xxhash128(ETag) is 32 hex chars and reuses the existing
`Document.content_hash` column per @yingfeng's suggestion; the connector
decides at listing time whether to populate it. Local uploads and
connectors that don't opt in fall through to the existing post-download
`xxhash128(blob)` path with no behavior change.
This is PR-1 of a 4-PR series — full design lives on #14628. Subsequent
PRs extend tier 1 to local FS / WebDAV / Dropbox / Seafile / RDBMS
(PR-2), wire up tier 2 cursor connectors with `SyncLogs.next_checkpoint`
(PR-3), and unify deletion via `KeyRecord(deleted=True)` reconciliation
(PR-4). Holding those back keeps this PR additive and reviewable on its
own.
#### Files touched
- `common/data_source/models.py` — new `KeyRecord`; optional
`fingerprint` on `Document`
- `common/data_source/interfaces.py` — `IncrementalCapability` enum,
`FingerprintConnector` ABC
- `common/data_source/blob_connector.py` — `BlobStorageConnector`
implements `FingerprintConnector`; per-object download factored into
`_build_document_from_obj()` so `_yield_blob_objects`, `list_keys`,
`get_value` all share it
- `rag/svr/sync_data_source.py` —
`_BlobLikeBase._fingerprint_filtered_generator` does the bypass loop;
`_run_task_logic` plumbs `doc.fingerprint` into the upload dict
- `api/db/services/document_service.py` —
`list_id_content_hash_map_by_kb_and_source_type()` helper
- `api/db/services/connector_service.py` + `file_service.py` —
fingerprint flows through `duplicate_and_parse → upload_document` and
lands in `content_hash`
- `test/unit_test/common/test_blob_connector_fingerprint.py` — 14 tests
covering ETag normalization (single-part, multipart, quoted, empty),
`list_keys()` not calling `GetObject`, `get_value()` materializing with
fingerprint, deterministic/stable fingerprints, and the bypass loop
asserting `GetObject` is *not* called on a match
#### Worth flagging for review
Old `_BlobLikeBase._generate` called `poll_source(start, now)` with a
`LastModified` window when `poll_range_start` was set. New code uses
`_fingerprint_filtered_generator` (full bucket listing + fingerprint
compare) outside of explicit `reindex=1`. Strictly better for
unchanged-bucket cases since it skips `GetObject`, but it does mean
every sync now does a full `list_objects_v2` paginate. Should still be
cheap for most buckets — flagging in case anyone has a very large bucket
where the time-window filter was meaningful.
On migration: existing rows have `content_hash = xxhash128(blob)` from
the old code. The first sync after this lands sees ETag-derived
fingerprints that don't match, re-fetches every object once, and writes
the new fingerprint. From the second sync onward the bypass works as
expected. "Slow day one, fast every day after." A `fingerprint_backfill:
trust` opt-out is sketched in the design doc but not in this PR.
#### Test plan
- [x] `uv run ruff check` — clean on all 8 touched files
- [x] `uv run pytest
test/unit_test/common/test_blob_connector_fingerprint.py -v` — 14 passed
- [x] Broader unit-test suite — no regressions in anything I touched
- [ ] Manual smoke against a real S3 bucket — configure a connector, run
sync twice, expect the second sync to log `bypassed=N, fetched=0` and no
`GetObject` calls in CloudTrail / bucket access logs
- [ ] Manual smoke with `reindex=1` — confirm the full re-download path
still works
### Type of change
- [x] New Feature (non-breaking change which adds functionality)
---------
Co-authored-by: Yingfeng <yingfeng.zhang@gmail.com>
2026-05-09 05:03:56 -07:00
from collections . abc import Iterator
2025-11-03 19:59:18 +08:00
from datetime import datetime , timezone
from typing import Any , Optional
feat(connectors): ETag-based bypass for incremental S3 ingestion (#14628) (#14677)
### What problem does this PR solve?
S3-family connector syncs currently re-download every in-window object
just so we can compute `xxhash128(blob)` and compare against
`Document.content_hash`. Anything that bumps `LastModified` without
changing bytes (`aws s3 cp` touches, bucket re-encryption, etc.) pays
full bandwidth and re-parses files that didn't actually change. #14628
covers the broader incremental-ingestion redesign; this PR is the first
slice.
The fix is a pre-listing short-circuit. `BlobStorageConnector` (S3 / R2
/ GCS / OCI / S3-compat) now implements a new `FingerprintConnector`
interface: `list_keys()` paginates `list_objects_v2` and yields
`KeyRecord(key, fingerprint)` where `fingerprint = xxhash128(ETag)`. The
orchestrator joins those against the connector's existing `{doc_id:
content_hash}` map and only calls `get_value(key)` when the fingerprint
differs. Unchanged keys are skipped entirely — no `GetObject`, no
re-parse.
No DDL. xxhash128(ETag) is 32 hex chars and reuses the existing
`Document.content_hash` column per @yingfeng's suggestion; the connector
decides at listing time whether to populate it. Local uploads and
connectors that don't opt in fall through to the existing post-download
`xxhash128(blob)` path with no behavior change.
This is PR-1 of a 4-PR series — full design lives on #14628. Subsequent
PRs extend tier 1 to local FS / WebDAV / Dropbox / Seafile / RDBMS
(PR-2), wire up tier 2 cursor connectors with `SyncLogs.next_checkpoint`
(PR-3), and unify deletion via `KeyRecord(deleted=True)` reconciliation
(PR-4). Holding those back keeps this PR additive and reviewable on its
own.
#### Files touched
- `common/data_source/models.py` — new `KeyRecord`; optional
`fingerprint` on `Document`
- `common/data_source/interfaces.py` — `IncrementalCapability` enum,
`FingerprintConnector` ABC
- `common/data_source/blob_connector.py` — `BlobStorageConnector`
implements `FingerprintConnector`; per-object download factored into
`_build_document_from_obj()` so `_yield_blob_objects`, `list_keys`,
`get_value` all share it
- `rag/svr/sync_data_source.py` —
`_BlobLikeBase._fingerprint_filtered_generator` does the bypass loop;
`_run_task_logic` plumbs `doc.fingerprint` into the upload dict
- `api/db/services/document_service.py` —
`list_id_content_hash_map_by_kb_and_source_type()` helper
- `api/db/services/connector_service.py` + `file_service.py` —
fingerprint flows through `duplicate_and_parse → upload_document` and
lands in `content_hash`
- `test/unit_test/common/test_blob_connector_fingerprint.py` — 14 tests
covering ETag normalization (single-part, multipart, quoted, empty),
`list_keys()` not calling `GetObject`, `get_value()` materializing with
fingerprint, deterministic/stable fingerprints, and the bypass loop
asserting `GetObject` is *not* called on a match
#### Worth flagging for review
Old `_BlobLikeBase._generate` called `poll_source(start, now)` with a
`LastModified` window when `poll_range_start` was set. New code uses
`_fingerprint_filtered_generator` (full bucket listing + fingerprint
compare) outside of explicit `reindex=1`. Strictly better for
unchanged-bucket cases since it skips `GetObject`, but it does mean
every sync now does a full `list_objects_v2` paginate. Should still be
cheap for most buckets — flagging in case anyone has a very large bucket
where the time-window filter was meaningful.
On migration: existing rows have `content_hash = xxhash128(blob)` from
the old code. The first sync after this lands sees ETag-derived
fingerprints that don't match, re-fetches every object once, and writes
the new fingerprint. From the second sync onward the bypass works as
expected. "Slow day one, fast every day after." A `fingerprint_backfill:
trust` opt-out is sketched in the design doc but not in this PR.
#### Test plan
- [x] `uv run ruff check` — clean on all 8 touched files
- [x] `uv run pytest
test/unit_test/common/test_blob_connector_fingerprint.py -v` — 14 passed
- [x] Broader unit-test suite — no regressions in anything I touched
- [ ] Manual smoke against a real S3 bucket — configure a connector, run
sync twice, expect the second sync to log `bypassed=N, fetched=0` and no
`GetObject` calls in CloudTrail / bucket access logs
- [ ] Manual smoke with `reindex=1` — confirm the full re-download path
still works
### Type of change
- [x] New Feature (non-breaking change which adds functionality)
---------
Co-authored-by: Yingfeng <yingfeng.zhang@gmail.com>
2026-05-09 05:03:56 -07:00
import xxhash
2025-11-03 19:59:18 +08:00
from common . data_source . utils import (
create_s3_client ,
detect_bucket_region ,
download_object ,
extract_size_bytes ,
get_file_ext ,
)
from common . data_source . config import BlobType , DocumentSource , BLOB_STORAGE_SIZE_THRESHOLD , INDEX_BATCH_SIZE
2026-07-03 12:53:39 +08:00
from common . data_source . exceptions import ConnectorMissingCredentialError , ConnectorValidationError , CredentialExpiredError , InsufficientPermissionsError
feat(connectors): ETag-based bypass for incremental S3 ingestion (#14628) (#14677)
### What problem does this PR solve?
S3-family connector syncs currently re-download every in-window object
just so we can compute `xxhash128(blob)` and compare against
`Document.content_hash`. Anything that bumps `LastModified` without
changing bytes (`aws s3 cp` touches, bucket re-encryption, etc.) pays
full bandwidth and re-parses files that didn't actually change. #14628
covers the broader incremental-ingestion redesign; this PR is the first
slice.
The fix is a pre-listing short-circuit. `BlobStorageConnector` (S3 / R2
/ GCS / OCI / S3-compat) now implements a new `FingerprintConnector`
interface: `list_keys()` paginates `list_objects_v2` and yields
`KeyRecord(key, fingerprint)` where `fingerprint = xxhash128(ETag)`. The
orchestrator joins those against the connector's existing `{doc_id:
content_hash}` map and only calls `get_value(key)` when the fingerprint
differs. Unchanged keys are skipped entirely — no `GetObject`, no
re-parse.
No DDL. xxhash128(ETag) is 32 hex chars and reuses the existing
`Document.content_hash` column per @yingfeng's suggestion; the connector
decides at listing time whether to populate it. Local uploads and
connectors that don't opt in fall through to the existing post-download
`xxhash128(blob)` path with no behavior change.
This is PR-1 of a 4-PR series — full design lives on #14628. Subsequent
PRs extend tier 1 to local FS / WebDAV / Dropbox / Seafile / RDBMS
(PR-2), wire up tier 2 cursor connectors with `SyncLogs.next_checkpoint`
(PR-3), and unify deletion via `KeyRecord(deleted=True)` reconciliation
(PR-4). Holding those back keeps this PR additive and reviewable on its
own.
#### Files touched
- `common/data_source/models.py` — new `KeyRecord`; optional
`fingerprint` on `Document`
- `common/data_source/interfaces.py` — `IncrementalCapability` enum,
`FingerprintConnector` ABC
- `common/data_source/blob_connector.py` — `BlobStorageConnector`
implements `FingerprintConnector`; per-object download factored into
`_build_document_from_obj()` so `_yield_blob_objects`, `list_keys`,
`get_value` all share it
- `rag/svr/sync_data_source.py` —
`_BlobLikeBase._fingerprint_filtered_generator` does the bypass loop;
`_run_task_logic` plumbs `doc.fingerprint` into the upload dict
- `api/db/services/document_service.py` —
`list_id_content_hash_map_by_kb_and_source_type()` helper
- `api/db/services/connector_service.py` + `file_service.py` —
fingerprint flows through `duplicate_and_parse → upload_document` and
lands in `content_hash`
- `test/unit_test/common/test_blob_connector_fingerprint.py` — 14 tests
covering ETag normalization (single-part, multipart, quoted, empty),
`list_keys()` not calling `GetObject`, `get_value()` materializing with
fingerprint, deterministic/stable fingerprints, and the bypass loop
asserting `GetObject` is *not* called on a match
#### Worth flagging for review
Old `_BlobLikeBase._generate` called `poll_source(start, now)` with a
`LastModified` window when `poll_range_start` was set. New code uses
`_fingerprint_filtered_generator` (full bucket listing + fingerprint
compare) outside of explicit `reindex=1`. Strictly better for
unchanged-bucket cases since it skips `GetObject`, but it does mean
every sync now does a full `list_objects_v2` paginate. Should still be
cheap for most buckets — flagging in case anyone has a very large bucket
where the time-window filter was meaningful.
On migration: existing rows have `content_hash = xxhash128(blob)` from
the old code. The first sync after this lands sees ETag-derived
fingerprints that don't match, re-fetches every object once, and writes
the new fingerprint. From the second sync onward the bypass works as
expected. "Slow day one, fast every day after." A `fingerprint_backfill:
trust` opt-out is sketched in the design doc but not in this PR.
#### Test plan
- [x] `uv run ruff check` — clean on all 8 touched files
- [x] `uv run pytest
test/unit_test/common/test_blob_connector_fingerprint.py -v` — 14 passed
- [x] Broader unit-test suite — no regressions in anything I touched
- [ ] Manual smoke against a real S3 bucket — configure a connector, run
sync twice, expect the second sync to log `bypassed=N, fetched=0` and no
`GetObject` calls in CloudTrail / bucket access logs
- [ ] Manual smoke with `reindex=1` — confirm the full re-download path
still works
### Type of change
- [x] New Feature (non-breaking change which adds functionality)
---------
Co-authored-by: Yingfeng <yingfeng.zhang@gmail.com>
2026-05-09 05:03:56 -07:00
from common . data_source . interfaces import (
FingerprintConnector ,
LoadConnector ,
PollConnector ,
)
2026-04-28 15:07:14 +08:00
from common . data_source . models import (
Document ,
feat(connectors): ETag-based bypass for incremental S3 ingestion (#14628) (#14677)
### What problem does this PR solve?
S3-family connector syncs currently re-download every in-window object
just so we can compute `xxhash128(blob)` and compare against
`Document.content_hash`. Anything that bumps `LastModified` without
changing bytes (`aws s3 cp` touches, bucket re-encryption, etc.) pays
full bandwidth and re-parses files that didn't actually change. #14628
covers the broader incremental-ingestion redesign; this PR is the first
slice.
The fix is a pre-listing short-circuit. `BlobStorageConnector` (S3 / R2
/ GCS / OCI / S3-compat) now implements a new `FingerprintConnector`
interface: `list_keys()` paginates `list_objects_v2` and yields
`KeyRecord(key, fingerprint)` where `fingerprint = xxhash128(ETag)`. The
orchestrator joins those against the connector's existing `{doc_id:
content_hash}` map and only calls `get_value(key)` when the fingerprint
differs. Unchanged keys are skipped entirely — no `GetObject`, no
re-parse.
No DDL. xxhash128(ETag) is 32 hex chars and reuses the existing
`Document.content_hash` column per @yingfeng's suggestion; the connector
decides at listing time whether to populate it. Local uploads and
connectors that don't opt in fall through to the existing post-download
`xxhash128(blob)` path with no behavior change.
This is PR-1 of a 4-PR series — full design lives on #14628. Subsequent
PRs extend tier 1 to local FS / WebDAV / Dropbox / Seafile / RDBMS
(PR-2), wire up tier 2 cursor connectors with `SyncLogs.next_checkpoint`
(PR-3), and unify deletion via `KeyRecord(deleted=True)` reconciliation
(PR-4). Holding those back keeps this PR additive and reviewable on its
own.
#### Files touched
- `common/data_source/models.py` — new `KeyRecord`; optional
`fingerprint` on `Document`
- `common/data_source/interfaces.py` — `IncrementalCapability` enum,
`FingerprintConnector` ABC
- `common/data_source/blob_connector.py` — `BlobStorageConnector`
implements `FingerprintConnector`; per-object download factored into
`_build_document_from_obj()` so `_yield_blob_objects`, `list_keys`,
`get_value` all share it
- `rag/svr/sync_data_source.py` —
`_BlobLikeBase._fingerprint_filtered_generator` does the bypass loop;
`_run_task_logic` plumbs `doc.fingerprint` into the upload dict
- `api/db/services/document_service.py` —
`list_id_content_hash_map_by_kb_and_source_type()` helper
- `api/db/services/connector_service.py` + `file_service.py` —
fingerprint flows through `duplicate_and_parse → upload_document` and
lands in `content_hash`
- `test/unit_test/common/test_blob_connector_fingerprint.py` — 14 tests
covering ETag normalization (single-part, multipart, quoted, empty),
`list_keys()` not calling `GetObject`, `get_value()` materializing with
fingerprint, deterministic/stable fingerprints, and the bypass loop
asserting `GetObject` is *not* called on a match
#### Worth flagging for review
Old `_BlobLikeBase._generate` called `poll_source(start, now)` with a
`LastModified` window when `poll_range_start` was set. New code uses
`_fingerprint_filtered_generator` (full bucket listing + fingerprint
compare) outside of explicit `reindex=1`. Strictly better for
unchanged-bucket cases since it skips `GetObject`, but it does mean
every sync now does a full `list_objects_v2` paginate. Should still be
cheap for most buckets — flagging in case anyone has a very large bucket
where the time-window filter was meaningful.
On migration: existing rows have `content_hash = xxhash128(blob)` from
the old code. The first sync after this lands sees ETag-derived
fingerprints that don't match, re-fetches every object once, and writes
the new fingerprint. From the second sync onward the bypass works as
expected. "Slow day one, fast every day after." A `fingerprint_backfill:
trust` opt-out is sketched in the design doc but not in this PR.
#### Test plan
- [x] `uv run ruff check` — clean on all 8 touched files
- [x] `uv run pytest
test/unit_test/common/test_blob_connector_fingerprint.py -v` — 14 passed
- [x] Broader unit-test suite — no regressions in anything I touched
- [ ] Manual smoke against a real S3 bucket — configure a connector, run
sync twice, expect the second sync to log `bypassed=N, fetched=0` and no
`GetObject` calls in CloudTrail / bucket access logs
- [ ] Manual smoke with `reindex=1` — confirm the full re-download path
still works
### Type of change
- [x] New Feature (non-breaking change which adds functionality)
---------
Co-authored-by: Yingfeng <yingfeng.zhang@gmail.com>
2026-05-09 05:03:56 -07:00
KeyRecord ,
2026-04-28 15:07:14 +08:00
SecondsSinceUnixEpoch ,
GenerateDocumentsOutput ,
GenerateSlimDocumentOutput ,
SlimDocument ,
)
2025-11-03 19:59:18 +08:00
feat(connectors): ETag-based bypass for incremental S3 ingestion (#14628) (#14677)
### What problem does this PR solve?
S3-family connector syncs currently re-download every in-window object
just so we can compute `xxhash128(blob)` and compare against
`Document.content_hash`. Anything that bumps `LastModified` without
changing bytes (`aws s3 cp` touches, bucket re-encryption, etc.) pays
full bandwidth and re-parses files that didn't actually change. #14628
covers the broader incremental-ingestion redesign; this PR is the first
slice.
The fix is a pre-listing short-circuit. `BlobStorageConnector` (S3 / R2
/ GCS / OCI / S3-compat) now implements a new `FingerprintConnector`
interface: `list_keys()` paginates `list_objects_v2` and yields
`KeyRecord(key, fingerprint)` where `fingerprint = xxhash128(ETag)`. The
orchestrator joins those against the connector's existing `{doc_id:
content_hash}` map and only calls `get_value(key)` when the fingerprint
differs. Unchanged keys are skipped entirely — no `GetObject`, no
re-parse.
No DDL. xxhash128(ETag) is 32 hex chars and reuses the existing
`Document.content_hash` column per @yingfeng's suggestion; the connector
decides at listing time whether to populate it. Local uploads and
connectors that don't opt in fall through to the existing post-download
`xxhash128(blob)` path with no behavior change.
This is PR-1 of a 4-PR series — full design lives on #14628. Subsequent
PRs extend tier 1 to local FS / WebDAV / Dropbox / Seafile / RDBMS
(PR-2), wire up tier 2 cursor connectors with `SyncLogs.next_checkpoint`
(PR-3), and unify deletion via `KeyRecord(deleted=True)` reconciliation
(PR-4). Holding those back keeps this PR additive and reviewable on its
own.
#### Files touched
- `common/data_source/models.py` — new `KeyRecord`; optional
`fingerprint` on `Document`
- `common/data_source/interfaces.py` — `IncrementalCapability` enum,
`FingerprintConnector` ABC
- `common/data_source/blob_connector.py` — `BlobStorageConnector`
implements `FingerprintConnector`; per-object download factored into
`_build_document_from_obj()` so `_yield_blob_objects`, `list_keys`,
`get_value` all share it
- `rag/svr/sync_data_source.py` —
`_BlobLikeBase._fingerprint_filtered_generator` does the bypass loop;
`_run_task_logic` plumbs `doc.fingerprint` into the upload dict
- `api/db/services/document_service.py` —
`list_id_content_hash_map_by_kb_and_source_type()` helper
- `api/db/services/connector_service.py` + `file_service.py` —
fingerprint flows through `duplicate_and_parse → upload_document` and
lands in `content_hash`
- `test/unit_test/common/test_blob_connector_fingerprint.py` — 14 tests
covering ETag normalization (single-part, multipart, quoted, empty),
`list_keys()` not calling `GetObject`, `get_value()` materializing with
fingerprint, deterministic/stable fingerprints, and the bypass loop
asserting `GetObject` is *not* called on a match
#### Worth flagging for review
Old `_BlobLikeBase._generate` called `poll_source(start, now)` with a
`LastModified` window when `poll_range_start` was set. New code uses
`_fingerprint_filtered_generator` (full bucket listing + fingerprint
compare) outside of explicit `reindex=1`. Strictly better for
unchanged-bucket cases since it skips `GetObject`, but it does mean
every sync now does a full `list_objects_v2` paginate. Should still be
cheap for most buckets — flagging in case anyone has a very large bucket
where the time-window filter was meaningful.
On migration: existing rows have `content_hash = xxhash128(blob)` from
the old code. The first sync after this lands sees ETag-derived
fingerprints that don't match, re-fetches every object once, and writes
the new fingerprint. From the second sync onward the bypass works as
expected. "Slow day one, fast every day after." A `fingerprint_backfill:
trust` opt-out is sketched in the design doc but not in this PR.
#### Test plan
- [x] `uv run ruff check` — clean on all 8 touched files
- [x] `uv run pytest
test/unit_test/common/test_blob_connector_fingerprint.py -v` — 14 passed
- [x] Broader unit-test suite — no regressions in anything I touched
- [ ] Manual smoke against a real S3 bucket — configure a connector, run
sync twice, expect the second sync to log `bypassed=N, fetched=0` and no
`GetObject` calls in CloudTrail / bucket access logs
- [ ] Manual smoke with `reindex=1` — confirm the full re-download path
still works
### Type of change
- [x] New Feature (non-breaking change which adds functionality)
---------
Co-authored-by: Yingfeng <yingfeng.zhang@gmail.com>
2026-05-09 05:03:56 -07:00
def _normalize_etag ( raw_etag : Optional [ str ] ) - > Optional [ str ] :
""" Return a 32-char hex fingerprint derived from an S3 ETag.
S3 ETags are MD5 ( 32 hex chars ) for single - part uploads and " <md5>-<n> "
( 34 + chars ) for multipart . We always hash so the column format is uniform
regardless of upload type or provider quirks ; equality of the hashed value
is sufficient for change detection .
"""
if not raw_etag :
return None
return xxhash . xxh128 ( raw_etag . strip ( ' " ' ) . encode ( ) ) . hexdigest ( )
class BlobStorageConnector ( LoadConnector , PollConnector , FingerprintConnector ) :
2025-11-03 19:59:18 +08:00
""" Blob storage connector """
def __init__ (
self ,
bucket_type : str ,
bucket_name : str ,
prefix : str = " " ,
batch_size : int = INDEX_BATCH_SIZE ,
european_residency : bool = False ,
) - > None :
self . bucket_type : BlobType = BlobType ( bucket_type )
self . bucket_name = bucket_name . strip ( )
self . prefix = prefix if not prefix or prefix . endswith ( " / " ) else prefix + " / "
self . batch_size = batch_size
self . s3_client : Optional [ Any ] = None
self . _allow_images : bool | None = None
self . size_threshold : int | None = BLOB_STORAGE_SIZE_THRESHOLD
self . bucket_region : Optional [ str ] = None
self . european_residency : bool = european_residency
feat(connectors): ETag-based bypass for incremental S3 ingestion (#14628) (#14677)
### What problem does this PR solve?
S3-family connector syncs currently re-download every in-window object
just so we can compute `xxhash128(blob)` and compare against
`Document.content_hash`. Anything that bumps `LastModified` without
changing bytes (`aws s3 cp` touches, bucket re-encryption, etc.) pays
full bandwidth and re-parses files that didn't actually change. #14628
covers the broader incremental-ingestion redesign; this PR is the first
slice.
The fix is a pre-listing short-circuit. `BlobStorageConnector` (S3 / R2
/ GCS / OCI / S3-compat) now implements a new `FingerprintConnector`
interface: `list_keys()` paginates `list_objects_v2` and yields
`KeyRecord(key, fingerprint)` where `fingerprint = xxhash128(ETag)`. The
orchestrator joins those against the connector's existing `{doc_id:
content_hash}` map and only calls `get_value(key)` when the fingerprint
differs. Unchanged keys are skipped entirely — no `GetObject`, no
re-parse.
No DDL. xxhash128(ETag) is 32 hex chars and reuses the existing
`Document.content_hash` column per @yingfeng's suggestion; the connector
decides at listing time whether to populate it. Local uploads and
connectors that don't opt in fall through to the existing post-download
`xxhash128(blob)` path with no behavior change.
This is PR-1 of a 4-PR series — full design lives on #14628. Subsequent
PRs extend tier 1 to local FS / WebDAV / Dropbox / Seafile / RDBMS
(PR-2), wire up tier 2 cursor connectors with `SyncLogs.next_checkpoint`
(PR-3), and unify deletion via `KeyRecord(deleted=True)` reconciliation
(PR-4). Holding those back keeps this PR additive and reviewable on its
own.
#### Files touched
- `common/data_source/models.py` — new `KeyRecord`; optional
`fingerprint` on `Document`
- `common/data_source/interfaces.py` — `IncrementalCapability` enum,
`FingerprintConnector` ABC
- `common/data_source/blob_connector.py` — `BlobStorageConnector`
implements `FingerprintConnector`; per-object download factored into
`_build_document_from_obj()` so `_yield_blob_objects`, `list_keys`,
`get_value` all share it
- `rag/svr/sync_data_source.py` —
`_BlobLikeBase._fingerprint_filtered_generator` does the bypass loop;
`_run_task_logic` plumbs `doc.fingerprint` into the upload dict
- `api/db/services/document_service.py` —
`list_id_content_hash_map_by_kb_and_source_type()` helper
- `api/db/services/connector_service.py` + `file_service.py` —
fingerprint flows through `duplicate_and_parse → upload_document` and
lands in `content_hash`
- `test/unit_test/common/test_blob_connector_fingerprint.py` — 14 tests
covering ETag normalization (single-part, multipart, quoted, empty),
`list_keys()` not calling `GetObject`, `get_value()` materializing with
fingerprint, deterministic/stable fingerprints, and the bypass loop
asserting `GetObject` is *not* called on a match
#### Worth flagging for review
Old `_BlobLikeBase._generate` called `poll_source(start, now)` with a
`LastModified` window when `poll_range_start` was set. New code uses
`_fingerprint_filtered_generator` (full bucket listing + fingerprint
compare) outside of explicit `reindex=1`. Strictly better for
unchanged-bucket cases since it skips `GetObject`, but it does mean
every sync now does a full `list_objects_v2` paginate. Should still be
cheap for most buckets — flagging in case anyone has a very large bucket
where the time-window filter was meaningful.
On migration: existing rows have `content_hash = xxhash128(blob)` from
the old code. The first sync after this lands sees ETag-derived
fingerprints that don't match, re-fetches every object once, and writes
the new fingerprint. From the second sync onward the bypass works as
expected. "Slow day one, fast every day after." A `fingerprint_backfill:
trust` opt-out is sketched in the design doc but not in this PR.
#### Test plan
- [x] `uv run ruff check` — clean on all 8 touched files
- [x] `uv run pytest
test/unit_test/common/test_blob_connector_fingerprint.py -v` — 14 passed
- [x] Broader unit-test suite — no regressions in anything I touched
- [ ] Manual smoke against a real S3 bucket — configure a connector, run
sync twice, expect the second sync to log `bypassed=N, fetched=0` and no
`GetObject` calls in CloudTrail / bucket access logs
- [ ] Manual smoke with `reindex=1` — confirm the full re-download path
still works
### Type of change
- [x] New Feature (non-breaking change which adds functionality)
---------
Co-authored-by: Yingfeng <yingfeng.zhang@gmail.com>
2026-05-09 05:03:56 -07:00
# Populated by list_keys() so a subsequent get_value(key) can find the
# raw S3 object metadata (LastModified, ETag, Key, Size) without a second
# head_object call. Lifetime is one list_keys() pass.
self . _listing_cache : dict [ str , dict [ str , Any ] ] = { }
self . _filename_counts : dict [ str , int ] = { }
2025-11-03 19:59:18 +08:00
def set_allow_images ( self , allow_images : bool ) - > None :
""" Set whether to process images """
logging . info ( f " Setting allow_images to { allow_images } . " )
self . _allow_images = allow_images
def load_credentials ( self , credentials : dict [ str , Any ] ) - > dict [ str , Any ] | None :
""" Load credentials """
2026-07-03 12:53:39 +08:00
logging . debug ( f " Loading credentials for { self . bucket_name } of type { self . bucket_type } " )
2025-11-03 19:59:18 +08:00
# Validate credentials
if self . bucket_type == BlobType . R2 :
2026-07-03 12:53:39 +08:00
if not all ( credentials . get ( key ) for key in [ " r2_access_key_id " , " r2_secret_access_key " , " account_id " ] ) :
2025-11-03 19:59:18 +08:00
raise ConnectorMissingCredentialError ( " Cloudflare R2 " )
elif self . bucket_type == BlobType . S3 :
authentication_method = credentials . get ( " authentication_method " , " access_key " )
2025-12-22 19:06:01 +08:00
2025-11-03 19:59:18 +08:00
if authentication_method == " access_key " :
2026-07-03 12:53:39 +08:00
if not all ( credentials . get ( key ) for key in [ " aws_access_key_id " , " aws_secret_access_key " ] ) :
2025-11-03 19:59:18 +08:00
raise ConnectorMissingCredentialError ( " Amazon S3 " )
2025-12-22 19:06:01 +08:00
2025-11-03 19:59:18 +08:00
elif authentication_method == " iam_role " :
if not credentials . get ( " aws_role_arn " ) :
raise ConnectorMissingCredentialError ( " Amazon S3 IAM role ARN is required " )
2026-07-03 12:53:39 +08:00
2025-12-22 19:06:01 +08:00
elif authentication_method == " assume_role " :
pass
else :
raise ConnectorMissingCredentialError ( " Unsupported S3 authentication method " )
2025-11-03 19:59:18 +08:00
elif self . bucket_type == BlobType . GOOGLE_CLOUD_STORAGE :
2026-07-03 12:53:39 +08:00
if not all ( credentials . get ( key ) for key in [ " access_key_id " , " secret_access_key " ] ) :
2025-11-03 19:59:18 +08:00
raise ConnectorMissingCredentialError ( " Google Cloud Storage " )
elif self . bucket_type == BlobType . OCI_STORAGE :
2026-07-03 12:53:39 +08:00
if not all ( credentials . get ( key ) for key in [ " namespace " , " region " , " access_key_id " , " secret_access_key " ] ) :
2025-11-03 19:59:18 +08:00
raise ConnectorMissingCredentialError ( " Oracle Cloud Infrastructure " )
2025-11-18 02:39:25 +01:00
elif self . bucket_type == BlobType . S3_COMPATIBLE :
2026-07-03 12:53:39 +08:00
if not all ( credentials . get ( key ) for key in [ " endpoint_url " , " aws_access_key_id " , " aws_secret_access_key " , " addressing_style " ] ) :
2025-11-18 02:39:25 +01:00
raise ConnectorMissingCredentialError ( " S3 Compatible Storage " )
2025-11-03 19:59:18 +08:00
else :
raise ValueError ( f " Unsupported bucket type: { self . bucket_type } " )
# Create S3 client
2026-07-03 12:53:39 +08:00
self . s3_client = create_s3_client ( self . bucket_type , credentials , self . european_residency )
2025-11-03 19:59:18 +08:00
# Detect bucket region (only important for S3)
if self . bucket_type == BlobType . S3 :
self . bucket_region = detect_bucket_region ( self . s3_client , self . bucket_name )
return None
feat(connectors): ETag-based bypass for incremental S3 ingestion (#14628) (#14677)
### What problem does this PR solve?
S3-family connector syncs currently re-download every in-window object
just so we can compute `xxhash128(blob)` and compare against
`Document.content_hash`. Anything that bumps `LastModified` without
changing bytes (`aws s3 cp` touches, bucket re-encryption, etc.) pays
full bandwidth and re-parses files that didn't actually change. #14628
covers the broader incremental-ingestion redesign; this PR is the first
slice.
The fix is a pre-listing short-circuit. `BlobStorageConnector` (S3 / R2
/ GCS / OCI / S3-compat) now implements a new `FingerprintConnector`
interface: `list_keys()` paginates `list_objects_v2` and yields
`KeyRecord(key, fingerprint)` where `fingerprint = xxhash128(ETag)`. The
orchestrator joins those against the connector's existing `{doc_id:
content_hash}` map and only calls `get_value(key)` when the fingerprint
differs. Unchanged keys are skipped entirely — no `GetObject`, no
re-parse.
No DDL. xxhash128(ETag) is 32 hex chars and reuses the existing
`Document.content_hash` column per @yingfeng's suggestion; the connector
decides at listing time whether to populate it. Local uploads and
connectors that don't opt in fall through to the existing post-download
`xxhash128(blob)` path with no behavior change.
This is PR-1 of a 4-PR series — full design lives on #14628. Subsequent
PRs extend tier 1 to local FS / WebDAV / Dropbox / Seafile / RDBMS
(PR-2), wire up tier 2 cursor connectors with `SyncLogs.next_checkpoint`
(PR-3), and unify deletion via `KeyRecord(deleted=True)` reconciliation
(PR-4). Holding those back keeps this PR additive and reviewable on its
own.
#### Files touched
- `common/data_source/models.py` — new `KeyRecord`; optional
`fingerprint` on `Document`
- `common/data_source/interfaces.py` — `IncrementalCapability` enum,
`FingerprintConnector` ABC
- `common/data_source/blob_connector.py` — `BlobStorageConnector`
implements `FingerprintConnector`; per-object download factored into
`_build_document_from_obj()` so `_yield_blob_objects`, `list_keys`,
`get_value` all share it
- `rag/svr/sync_data_source.py` —
`_BlobLikeBase._fingerprint_filtered_generator` does the bypass loop;
`_run_task_logic` plumbs `doc.fingerprint` into the upload dict
- `api/db/services/document_service.py` —
`list_id_content_hash_map_by_kb_and_source_type()` helper
- `api/db/services/connector_service.py` + `file_service.py` —
fingerprint flows through `duplicate_and_parse → upload_document` and
lands in `content_hash`
- `test/unit_test/common/test_blob_connector_fingerprint.py` — 14 tests
covering ETag normalization (single-part, multipart, quoted, empty),
`list_keys()` not calling `GetObject`, `get_value()` materializing with
fingerprint, deterministic/stable fingerprints, and the bypass loop
asserting `GetObject` is *not* called on a match
#### Worth flagging for review
Old `_BlobLikeBase._generate` called `poll_source(start, now)` with a
`LastModified` window when `poll_range_start` was set. New code uses
`_fingerprint_filtered_generator` (full bucket listing + fingerprint
compare) outside of explicit `reindex=1`. Strictly better for
unchanged-bucket cases since it skips `GetObject`, but it does mean
every sync now does a full `list_objects_v2` paginate. Should still be
cheap for most buckets — flagging in case anyone has a very large bucket
where the time-window filter was meaningful.
On migration: existing rows have `content_hash = xxhash128(blob)` from
the old code. The first sync after this lands sees ETag-derived
fingerprints that don't match, re-fetches every object once, and writes
the new fingerprint. From the second sync onward the bypass works as
expected. "Slow day one, fast every day after." A `fingerprint_backfill:
trust` opt-out is sketched in the design doc but not in this PR.
#### Test plan
- [x] `uv run ruff check` — clean on all 8 touched files
- [x] `uv run pytest
test/unit_test/common/test_blob_connector_fingerprint.py -v` — 14 passed
- [x] Broader unit-test suite — no regressions in anything I touched
- [ ] Manual smoke against a real S3 bucket — configure a connector, run
sync twice, expect the second sync to log `bypassed=N, fetched=0` and no
`GetObject` calls in CloudTrail / bucket access logs
- [ ] Manual smoke with `reindex=1` — confirm the full re-download path
still works
### Type of change
- [x] New Feature (non-breaking change which adds functionality)
---------
Co-authored-by: Yingfeng <yingfeng.zhang@gmail.com>
2026-05-09 05:03:56 -07:00
def _build_document_from_obj (
self ,
obj : dict [ str , Any ] ,
filename_counts : dict [ str , int ] ,
) - > Optional [ Document ] :
""" Materialize a Document for one S3 object, downloading its body. """
key = obj [ " Key " ]
file_name = os . path . basename ( key )
last_modified = obj [ " LastModified " ] . replace ( tzinfo = timezone . utc )
size_bytes = extract_size_bytes ( obj )
2026-07-03 12:53:39 +08:00
if self . size_threshold is not None and isinstance ( size_bytes , int ) and size_bytes > self . size_threshold :
logging . warning ( f " { file_name } exceeds size threshold of { self . size_threshold } . Skipping. " )
feat(connectors): ETag-based bypass for incremental S3 ingestion (#14628) (#14677)
### What problem does this PR solve?
S3-family connector syncs currently re-download every in-window object
just so we can compute `xxhash128(blob)` and compare against
`Document.content_hash`. Anything that bumps `LastModified` without
changing bytes (`aws s3 cp` touches, bucket re-encryption, etc.) pays
full bandwidth and re-parses files that didn't actually change. #14628
covers the broader incremental-ingestion redesign; this PR is the first
slice.
The fix is a pre-listing short-circuit. `BlobStorageConnector` (S3 / R2
/ GCS / OCI / S3-compat) now implements a new `FingerprintConnector`
interface: `list_keys()` paginates `list_objects_v2` and yields
`KeyRecord(key, fingerprint)` where `fingerprint = xxhash128(ETag)`. The
orchestrator joins those against the connector's existing `{doc_id:
content_hash}` map and only calls `get_value(key)` when the fingerprint
differs. Unchanged keys are skipped entirely — no `GetObject`, no
re-parse.
No DDL. xxhash128(ETag) is 32 hex chars and reuses the existing
`Document.content_hash` column per @yingfeng's suggestion; the connector
decides at listing time whether to populate it. Local uploads and
connectors that don't opt in fall through to the existing post-download
`xxhash128(blob)` path with no behavior change.
This is PR-1 of a 4-PR series — full design lives on #14628. Subsequent
PRs extend tier 1 to local FS / WebDAV / Dropbox / Seafile / RDBMS
(PR-2), wire up tier 2 cursor connectors with `SyncLogs.next_checkpoint`
(PR-3), and unify deletion via `KeyRecord(deleted=True)` reconciliation
(PR-4). Holding those back keeps this PR additive and reviewable on its
own.
#### Files touched
- `common/data_source/models.py` — new `KeyRecord`; optional
`fingerprint` on `Document`
- `common/data_source/interfaces.py` — `IncrementalCapability` enum,
`FingerprintConnector` ABC
- `common/data_source/blob_connector.py` — `BlobStorageConnector`
implements `FingerprintConnector`; per-object download factored into
`_build_document_from_obj()` so `_yield_blob_objects`, `list_keys`,
`get_value` all share it
- `rag/svr/sync_data_source.py` —
`_BlobLikeBase._fingerprint_filtered_generator` does the bypass loop;
`_run_task_logic` plumbs `doc.fingerprint` into the upload dict
- `api/db/services/document_service.py` —
`list_id_content_hash_map_by_kb_and_source_type()` helper
- `api/db/services/connector_service.py` + `file_service.py` —
fingerprint flows through `duplicate_and_parse → upload_document` and
lands in `content_hash`
- `test/unit_test/common/test_blob_connector_fingerprint.py` — 14 tests
covering ETag normalization (single-part, multipart, quoted, empty),
`list_keys()` not calling `GetObject`, `get_value()` materializing with
fingerprint, deterministic/stable fingerprints, and the bypass loop
asserting `GetObject` is *not* called on a match
#### Worth flagging for review
Old `_BlobLikeBase._generate` called `poll_source(start, now)` with a
`LastModified` window when `poll_range_start` was set. New code uses
`_fingerprint_filtered_generator` (full bucket listing + fingerprint
compare) outside of explicit `reindex=1`. Strictly better for
unchanged-bucket cases since it skips `GetObject`, but it does mean
every sync now does a full `list_objects_v2` paginate. Should still be
cheap for most buckets — flagging in case anyone has a very large bucket
where the time-window filter was meaningful.
On migration: existing rows have `content_hash = xxhash128(blob)` from
the old code. The first sync after this lands sees ETag-derived
fingerprints that don't match, re-fetches every object once, and writes
the new fingerprint. From the second sync onward the bypass works as
expected. "Slow day one, fast every day after." A `fingerprint_backfill:
trust` opt-out is sketched in the design doc but not in this PR.
#### Test plan
- [x] `uv run ruff check` — clean on all 8 touched files
- [x] `uv run pytest
test/unit_test/common/test_blob_connector_fingerprint.py -v` — 14 passed
- [x] Broader unit-test suite — no regressions in anything I touched
- [ ] Manual smoke against a real S3 bucket — configure a connector, run
sync twice, expect the second sync to log `bypassed=N, fetched=0` and no
`GetObject` calls in CloudTrail / bucket access logs
- [ ] Manual smoke with `reindex=1` — confirm the full re-download path
still works
### Type of change
- [x] New Feature (non-breaking change which adds functionality)
---------
Co-authored-by: Yingfeng <yingfeng.zhang@gmail.com>
2026-05-09 05:03:56 -07:00
return None
2026-07-03 12:53:39 +08:00
blob = download_object ( self . s3_client , self . bucket_name , key , self . size_threshold )
feat(connectors): ETag-based bypass for incremental S3 ingestion (#14628) (#14677)
### What problem does this PR solve?
S3-family connector syncs currently re-download every in-window object
just so we can compute `xxhash128(blob)` and compare against
`Document.content_hash`. Anything that bumps `LastModified` without
changing bytes (`aws s3 cp` touches, bucket re-encryption, etc.) pays
full bandwidth and re-parses files that didn't actually change. #14628
covers the broader incremental-ingestion redesign; this PR is the first
slice.
The fix is a pre-listing short-circuit. `BlobStorageConnector` (S3 / R2
/ GCS / OCI / S3-compat) now implements a new `FingerprintConnector`
interface: `list_keys()` paginates `list_objects_v2` and yields
`KeyRecord(key, fingerprint)` where `fingerprint = xxhash128(ETag)`. The
orchestrator joins those against the connector's existing `{doc_id:
content_hash}` map and only calls `get_value(key)` when the fingerprint
differs. Unchanged keys are skipped entirely — no `GetObject`, no
re-parse.
No DDL. xxhash128(ETag) is 32 hex chars and reuses the existing
`Document.content_hash` column per @yingfeng's suggestion; the connector
decides at listing time whether to populate it. Local uploads and
connectors that don't opt in fall through to the existing post-download
`xxhash128(blob)` path with no behavior change.
This is PR-1 of a 4-PR series — full design lives on #14628. Subsequent
PRs extend tier 1 to local FS / WebDAV / Dropbox / Seafile / RDBMS
(PR-2), wire up tier 2 cursor connectors with `SyncLogs.next_checkpoint`
(PR-3), and unify deletion via `KeyRecord(deleted=True)` reconciliation
(PR-4). Holding those back keeps this PR additive and reviewable on its
own.
#### Files touched
- `common/data_source/models.py` — new `KeyRecord`; optional
`fingerprint` on `Document`
- `common/data_source/interfaces.py` — `IncrementalCapability` enum,
`FingerprintConnector` ABC
- `common/data_source/blob_connector.py` — `BlobStorageConnector`
implements `FingerprintConnector`; per-object download factored into
`_build_document_from_obj()` so `_yield_blob_objects`, `list_keys`,
`get_value` all share it
- `rag/svr/sync_data_source.py` —
`_BlobLikeBase._fingerprint_filtered_generator` does the bypass loop;
`_run_task_logic` plumbs `doc.fingerprint` into the upload dict
- `api/db/services/document_service.py` —
`list_id_content_hash_map_by_kb_and_source_type()` helper
- `api/db/services/connector_service.py` + `file_service.py` —
fingerprint flows through `duplicate_and_parse → upload_document` and
lands in `content_hash`
- `test/unit_test/common/test_blob_connector_fingerprint.py` — 14 tests
covering ETag normalization (single-part, multipart, quoted, empty),
`list_keys()` not calling `GetObject`, `get_value()` materializing with
fingerprint, deterministic/stable fingerprints, and the bypass loop
asserting `GetObject` is *not* called on a match
#### Worth flagging for review
Old `_BlobLikeBase._generate` called `poll_source(start, now)` with a
`LastModified` window when `poll_range_start` was set. New code uses
`_fingerprint_filtered_generator` (full bucket listing + fingerprint
compare) outside of explicit `reindex=1`. Strictly better for
unchanged-bucket cases since it skips `GetObject`, but it does mean
every sync now does a full `list_objects_v2` paginate. Should still be
cheap for most buckets — flagging in case anyone has a very large bucket
where the time-window filter was meaningful.
On migration: existing rows have `content_hash = xxhash128(blob)` from
the old code. The first sync after this lands sees ETag-derived
fingerprints that don't match, re-fetches every object once, and writes
the new fingerprint. From the second sync onward the bypass works as
expected. "Slow day one, fast every day after." A `fingerprint_backfill:
trust` opt-out is sketched in the design doc but not in this PR.
#### Test plan
- [x] `uv run ruff check` — clean on all 8 touched files
- [x] `uv run pytest
test/unit_test/common/test_blob_connector_fingerprint.py -v` — 14 passed
- [x] Broader unit-test suite — no regressions in anything I touched
- [ ] Manual smoke against a real S3 bucket — configure a connector, run
sync twice, expect the second sync to log `bypassed=N, fetched=0` and no
`GetObject` calls in CloudTrail / bucket access logs
- [ ] Manual smoke with `reindex=1` — confirm the full re-download path
still works
### Type of change
- [x] New Feature (non-breaking change which adds functionality)
---------
Co-authored-by: Yingfeng <yingfeng.zhang@gmail.com>
2026-05-09 05:03:56 -07:00
if blob is None :
return None
return Document (
id = f " { self . bucket_type } : { self . bucket_name } : { key } " ,
blob = blob ,
source = DocumentSource ( self . bucket_type . value ) ,
semantic_identifier = self . _get_semantic_id ( key , file_name , filename_counts ) ,
extension = get_file_ext ( file_name ) ,
doc_updated_at = last_modified ,
size_bytes = size_bytes if size_bytes else 0 ,
fingerprint = _normalize_etag ( obj . get ( " ETag " ) ) ,
)
2025-11-03 19:59:18 +08:00
def _yield_blob_objects (
self ,
start : datetime ,
end : datetime ,
) - > GenerateDocumentsOutput :
""" Generate bucket objects """
2026-04-28 15:07:14 +08:00
all_objects , filename_counts = self . _collect_blob_objects ( start , end )
2025-11-03 19:59:18 +08:00
2025-12-18 10:42:43 +01:00
batch : list [ Document ] = [ ]
for obj in all_objects :
try :
feat(connectors): ETag-based bypass for incremental S3 ingestion (#14628) (#14677)
### What problem does this PR solve?
S3-family connector syncs currently re-download every in-window object
just so we can compute `xxhash128(blob)` and compare against
`Document.content_hash`. Anything that bumps `LastModified` without
changing bytes (`aws s3 cp` touches, bucket re-encryption, etc.) pays
full bandwidth and re-parses files that didn't actually change. #14628
covers the broader incremental-ingestion redesign; this PR is the first
slice.
The fix is a pre-listing short-circuit. `BlobStorageConnector` (S3 / R2
/ GCS / OCI / S3-compat) now implements a new `FingerprintConnector`
interface: `list_keys()` paginates `list_objects_v2` and yields
`KeyRecord(key, fingerprint)` where `fingerprint = xxhash128(ETag)`. The
orchestrator joins those against the connector's existing `{doc_id:
content_hash}` map and only calls `get_value(key)` when the fingerprint
differs. Unchanged keys are skipped entirely — no `GetObject`, no
re-parse.
No DDL. xxhash128(ETag) is 32 hex chars and reuses the existing
`Document.content_hash` column per @yingfeng's suggestion; the connector
decides at listing time whether to populate it. Local uploads and
connectors that don't opt in fall through to the existing post-download
`xxhash128(blob)` path with no behavior change.
This is PR-1 of a 4-PR series — full design lives on #14628. Subsequent
PRs extend tier 1 to local FS / WebDAV / Dropbox / Seafile / RDBMS
(PR-2), wire up tier 2 cursor connectors with `SyncLogs.next_checkpoint`
(PR-3), and unify deletion via `KeyRecord(deleted=True)` reconciliation
(PR-4). Holding those back keeps this PR additive and reviewable on its
own.
#### Files touched
- `common/data_source/models.py` — new `KeyRecord`; optional
`fingerprint` on `Document`
- `common/data_source/interfaces.py` — `IncrementalCapability` enum,
`FingerprintConnector` ABC
- `common/data_source/blob_connector.py` — `BlobStorageConnector`
implements `FingerprintConnector`; per-object download factored into
`_build_document_from_obj()` so `_yield_blob_objects`, `list_keys`,
`get_value` all share it
- `rag/svr/sync_data_source.py` —
`_BlobLikeBase._fingerprint_filtered_generator` does the bypass loop;
`_run_task_logic` plumbs `doc.fingerprint` into the upload dict
- `api/db/services/document_service.py` —
`list_id_content_hash_map_by_kb_and_source_type()` helper
- `api/db/services/connector_service.py` + `file_service.py` —
fingerprint flows through `duplicate_and_parse → upload_document` and
lands in `content_hash`
- `test/unit_test/common/test_blob_connector_fingerprint.py` — 14 tests
covering ETag normalization (single-part, multipart, quoted, empty),
`list_keys()` not calling `GetObject`, `get_value()` materializing with
fingerprint, deterministic/stable fingerprints, and the bypass loop
asserting `GetObject` is *not* called on a match
#### Worth flagging for review
Old `_BlobLikeBase._generate` called `poll_source(start, now)` with a
`LastModified` window when `poll_range_start` was set. New code uses
`_fingerprint_filtered_generator` (full bucket listing + fingerprint
compare) outside of explicit `reindex=1`. Strictly better for
unchanged-bucket cases since it skips `GetObject`, but it does mean
every sync now does a full `list_objects_v2` paginate. Should still be
cheap for most buckets — flagging in case anyone has a very large bucket
where the time-window filter was meaningful.
On migration: existing rows have `content_hash = xxhash128(blob)` from
the old code. The first sync after this lands sees ETag-derived
fingerprints that don't match, re-fetches every object once, and writes
the new fingerprint. From the second sync onward the bypass works as
expected. "Slow day one, fast every day after." A `fingerprint_backfill:
trust` opt-out is sketched in the design doc but not in this PR.
#### Test plan
- [x] `uv run ruff check` — clean on all 8 touched files
- [x] `uv run pytest
test/unit_test/common/test_blob_connector_fingerprint.py -v` — 14 passed
- [x] Broader unit-test suite — no regressions in anything I touched
- [ ] Manual smoke against a real S3 bucket — configure a connector, run
sync twice, expect the second sync to log `bypassed=N, fetched=0` and no
`GetObject` calls in CloudTrail / bucket access logs
- [ ] Manual smoke with `reindex=1` — confirm the full re-download path
still works
### Type of change
- [x] New Feature (non-breaking change which adds functionality)
---------
Co-authored-by: Yingfeng <yingfeng.zhang@gmail.com>
2026-05-09 05:03:56 -07:00
doc = self . _build_document_from_obj ( obj , filename_counts )
if doc is None :
2025-11-03 19:59:18 +08:00
continue
feat(connectors): ETag-based bypass for incremental S3 ingestion (#14628) (#14677)
### What problem does this PR solve?
S3-family connector syncs currently re-download every in-window object
just so we can compute `xxhash128(blob)` and compare against
`Document.content_hash`. Anything that bumps `LastModified` without
changing bytes (`aws s3 cp` touches, bucket re-encryption, etc.) pays
full bandwidth and re-parses files that didn't actually change. #14628
covers the broader incremental-ingestion redesign; this PR is the first
slice.
The fix is a pre-listing short-circuit. `BlobStorageConnector` (S3 / R2
/ GCS / OCI / S3-compat) now implements a new `FingerprintConnector`
interface: `list_keys()` paginates `list_objects_v2` and yields
`KeyRecord(key, fingerprint)` where `fingerprint = xxhash128(ETag)`. The
orchestrator joins those against the connector's existing `{doc_id:
content_hash}` map and only calls `get_value(key)` when the fingerprint
differs. Unchanged keys are skipped entirely — no `GetObject`, no
re-parse.
No DDL. xxhash128(ETag) is 32 hex chars and reuses the existing
`Document.content_hash` column per @yingfeng's suggestion; the connector
decides at listing time whether to populate it. Local uploads and
connectors that don't opt in fall through to the existing post-download
`xxhash128(blob)` path with no behavior change.
This is PR-1 of a 4-PR series — full design lives on #14628. Subsequent
PRs extend tier 1 to local FS / WebDAV / Dropbox / Seafile / RDBMS
(PR-2), wire up tier 2 cursor connectors with `SyncLogs.next_checkpoint`
(PR-3), and unify deletion via `KeyRecord(deleted=True)` reconciliation
(PR-4). Holding those back keeps this PR additive and reviewable on its
own.
#### Files touched
- `common/data_source/models.py` — new `KeyRecord`; optional
`fingerprint` on `Document`
- `common/data_source/interfaces.py` — `IncrementalCapability` enum,
`FingerprintConnector` ABC
- `common/data_source/blob_connector.py` — `BlobStorageConnector`
implements `FingerprintConnector`; per-object download factored into
`_build_document_from_obj()` so `_yield_blob_objects`, `list_keys`,
`get_value` all share it
- `rag/svr/sync_data_source.py` —
`_BlobLikeBase._fingerprint_filtered_generator` does the bypass loop;
`_run_task_logic` plumbs `doc.fingerprint` into the upload dict
- `api/db/services/document_service.py` —
`list_id_content_hash_map_by_kb_and_source_type()` helper
- `api/db/services/connector_service.py` + `file_service.py` —
fingerprint flows through `duplicate_and_parse → upload_document` and
lands in `content_hash`
- `test/unit_test/common/test_blob_connector_fingerprint.py` — 14 tests
covering ETag normalization (single-part, multipart, quoted, empty),
`list_keys()` not calling `GetObject`, `get_value()` materializing with
fingerprint, deterministic/stable fingerprints, and the bypass loop
asserting `GetObject` is *not* called on a match
#### Worth flagging for review
Old `_BlobLikeBase._generate` called `poll_source(start, now)` with a
`LastModified` window when `poll_range_start` was set. New code uses
`_fingerprint_filtered_generator` (full bucket listing + fingerprint
compare) outside of explicit `reindex=1`. Strictly better for
unchanged-bucket cases since it skips `GetObject`, but it does mean
every sync now does a full `list_objects_v2` paginate. Should still be
cheap for most buckets — flagging in case anyone has a very large bucket
where the time-window filter was meaningful.
On migration: existing rows have `content_hash = xxhash128(blob)` from
the old code. The first sync after this lands sees ETag-derived
fingerprints that don't match, re-fetches every object once, and writes
the new fingerprint. From the second sync onward the bypass works as
expected. "Slow day one, fast every day after." A `fingerprint_backfill:
trust` opt-out is sketched in the design doc but not in this PR.
#### Test plan
- [x] `uv run ruff check` — clean on all 8 touched files
- [x] `uv run pytest
test/unit_test/common/test_blob_connector_fingerprint.py -v` — 14 passed
- [x] Broader unit-test suite — no regressions in anything I touched
- [ ] Manual smoke against a real S3 bucket — configure a connector, run
sync twice, expect the second sync to log `bypassed=N, fetched=0` and no
`GetObject` calls in CloudTrail / bucket access logs
- [ ] Manual smoke with `reindex=1` — confirm the full re-download path
still works
### Type of change
- [x] New Feature (non-breaking change which adds functionality)
---------
Co-authored-by: Yingfeng <yingfeng.zhang@gmail.com>
2026-05-09 05:03:56 -07:00
batch . append ( doc )
2025-12-18 10:42:43 +01:00
if len ( batch ) == self . batch_size :
yield batch
batch = [ ]
except Exception :
feat(connectors): ETag-based bypass for incremental S3 ingestion (#14628) (#14677)
### What problem does this PR solve?
S3-family connector syncs currently re-download every in-window object
just so we can compute `xxhash128(blob)` and compare against
`Document.content_hash`. Anything that bumps `LastModified` without
changing bytes (`aws s3 cp` touches, bucket re-encryption, etc.) pays
full bandwidth and re-parses files that didn't actually change. #14628
covers the broader incremental-ingestion redesign; this PR is the first
slice.
The fix is a pre-listing short-circuit. `BlobStorageConnector` (S3 / R2
/ GCS / OCI / S3-compat) now implements a new `FingerprintConnector`
interface: `list_keys()` paginates `list_objects_v2` and yields
`KeyRecord(key, fingerprint)` where `fingerprint = xxhash128(ETag)`. The
orchestrator joins those against the connector's existing `{doc_id:
content_hash}` map and only calls `get_value(key)` when the fingerprint
differs. Unchanged keys are skipped entirely — no `GetObject`, no
re-parse.
No DDL. xxhash128(ETag) is 32 hex chars and reuses the existing
`Document.content_hash` column per @yingfeng's suggestion; the connector
decides at listing time whether to populate it. Local uploads and
connectors that don't opt in fall through to the existing post-download
`xxhash128(blob)` path with no behavior change.
This is PR-1 of a 4-PR series — full design lives on #14628. Subsequent
PRs extend tier 1 to local FS / WebDAV / Dropbox / Seafile / RDBMS
(PR-2), wire up tier 2 cursor connectors with `SyncLogs.next_checkpoint`
(PR-3), and unify deletion via `KeyRecord(deleted=True)` reconciliation
(PR-4). Holding those back keeps this PR additive and reviewable on its
own.
#### Files touched
- `common/data_source/models.py` — new `KeyRecord`; optional
`fingerprint` on `Document`
- `common/data_source/interfaces.py` — `IncrementalCapability` enum,
`FingerprintConnector` ABC
- `common/data_source/blob_connector.py` — `BlobStorageConnector`
implements `FingerprintConnector`; per-object download factored into
`_build_document_from_obj()` so `_yield_blob_objects`, `list_keys`,
`get_value` all share it
- `rag/svr/sync_data_source.py` —
`_BlobLikeBase._fingerprint_filtered_generator` does the bypass loop;
`_run_task_logic` plumbs `doc.fingerprint` into the upload dict
- `api/db/services/document_service.py` —
`list_id_content_hash_map_by_kb_and_source_type()` helper
- `api/db/services/connector_service.py` + `file_service.py` —
fingerprint flows through `duplicate_and_parse → upload_document` and
lands in `content_hash`
- `test/unit_test/common/test_blob_connector_fingerprint.py` — 14 tests
covering ETag normalization (single-part, multipart, quoted, empty),
`list_keys()` not calling `GetObject`, `get_value()` materializing with
fingerprint, deterministic/stable fingerprints, and the bypass loop
asserting `GetObject` is *not* called on a match
#### Worth flagging for review
Old `_BlobLikeBase._generate` called `poll_source(start, now)` with a
`LastModified` window when `poll_range_start` was set. New code uses
`_fingerprint_filtered_generator` (full bucket listing + fingerprint
compare) outside of explicit `reindex=1`. Strictly better for
unchanged-bucket cases since it skips `GetObject`, but it does mean
every sync now does a full `list_objects_v2` paginate. Should still be
cheap for most buckets — flagging in case anyone has a very large bucket
where the time-window filter was meaningful.
On migration: existing rows have `content_hash = xxhash128(blob)` from
the old code. The first sync after this lands sees ETag-derived
fingerprints that don't match, re-fetches every object once, and writes
the new fingerprint. From the second sync onward the bypass works as
expected. "Slow day one, fast every day after." A `fingerprint_backfill:
trust` opt-out is sketched in the design doc but not in this PR.
#### Test plan
- [x] `uv run ruff check` — clean on all 8 touched files
- [x] `uv run pytest
test/unit_test/common/test_blob_connector_fingerprint.py -v` — 14 passed
- [x] Broader unit-test suite — no regressions in anything I touched
- [ ] Manual smoke against a real S3 bucket — configure a connector, run
sync twice, expect the second sync to log `bypassed=N, fetched=0` and no
`GetObject` calls in CloudTrail / bucket access logs
- [ ] Manual smoke with `reindex=1` — confirm the full re-download path
still works
### Type of change
- [x] New Feature (non-breaking change which adds functionality)
---------
Co-authored-by: Yingfeng <yingfeng.zhang@gmail.com>
2026-05-09 05:03:56 -07:00
logging . exception ( f " Error decoding object { obj . get ( ' Key ' ) } " )
2026-04-28 15:07:14 +08:00
if batch :
yield batch
feat(connectors): ETag-based bypass for incremental S3 ingestion (#14628) (#14677)
### What problem does this PR solve?
S3-family connector syncs currently re-download every in-window object
just so we can compute `xxhash128(blob)` and compare against
`Document.content_hash`. Anything that bumps `LastModified` without
changing bytes (`aws s3 cp` touches, bucket re-encryption, etc.) pays
full bandwidth and re-parses files that didn't actually change. #14628
covers the broader incremental-ingestion redesign; this PR is the first
slice.
The fix is a pre-listing short-circuit. `BlobStorageConnector` (S3 / R2
/ GCS / OCI / S3-compat) now implements a new `FingerprintConnector`
interface: `list_keys()` paginates `list_objects_v2` and yields
`KeyRecord(key, fingerprint)` where `fingerprint = xxhash128(ETag)`. The
orchestrator joins those against the connector's existing `{doc_id:
content_hash}` map and only calls `get_value(key)` when the fingerprint
differs. Unchanged keys are skipped entirely — no `GetObject`, no
re-parse.
No DDL. xxhash128(ETag) is 32 hex chars and reuses the existing
`Document.content_hash` column per @yingfeng's suggestion; the connector
decides at listing time whether to populate it. Local uploads and
connectors that don't opt in fall through to the existing post-download
`xxhash128(blob)` path with no behavior change.
This is PR-1 of a 4-PR series — full design lives on #14628. Subsequent
PRs extend tier 1 to local FS / WebDAV / Dropbox / Seafile / RDBMS
(PR-2), wire up tier 2 cursor connectors with `SyncLogs.next_checkpoint`
(PR-3), and unify deletion via `KeyRecord(deleted=True)` reconciliation
(PR-4). Holding those back keeps this PR additive and reviewable on its
own.
#### Files touched
- `common/data_source/models.py` — new `KeyRecord`; optional
`fingerprint` on `Document`
- `common/data_source/interfaces.py` — `IncrementalCapability` enum,
`FingerprintConnector` ABC
- `common/data_source/blob_connector.py` — `BlobStorageConnector`
implements `FingerprintConnector`; per-object download factored into
`_build_document_from_obj()` so `_yield_blob_objects`, `list_keys`,
`get_value` all share it
- `rag/svr/sync_data_source.py` —
`_BlobLikeBase._fingerprint_filtered_generator` does the bypass loop;
`_run_task_logic` plumbs `doc.fingerprint` into the upload dict
- `api/db/services/document_service.py` —
`list_id_content_hash_map_by_kb_and_source_type()` helper
- `api/db/services/connector_service.py` + `file_service.py` —
fingerprint flows through `duplicate_and_parse → upload_document` and
lands in `content_hash`
- `test/unit_test/common/test_blob_connector_fingerprint.py` — 14 tests
covering ETag normalization (single-part, multipart, quoted, empty),
`list_keys()` not calling `GetObject`, `get_value()` materializing with
fingerprint, deterministic/stable fingerprints, and the bypass loop
asserting `GetObject` is *not* called on a match
#### Worth flagging for review
Old `_BlobLikeBase._generate` called `poll_source(start, now)` with a
`LastModified` window when `poll_range_start` was set. New code uses
`_fingerprint_filtered_generator` (full bucket listing + fingerprint
compare) outside of explicit `reindex=1`. Strictly better for
unchanged-bucket cases since it skips `GetObject`, but it does mean
every sync now does a full `list_objects_v2` paginate. Should still be
cheap for most buckets — flagging in case anyone has a very large bucket
where the time-window filter was meaningful.
On migration: existing rows have `content_hash = xxhash128(blob)` from
the old code. The first sync after this lands sees ETag-derived
fingerprints that don't match, re-fetches every object once, and writes
the new fingerprint. From the second sync onward the bypass works as
expected. "Slow day one, fast every day after." A `fingerprint_backfill:
trust` opt-out is sketched in the design doc but not in this PR.
#### Test plan
- [x] `uv run ruff check` — clean on all 8 touched files
- [x] `uv run pytest
test/unit_test/common/test_blob_connector_fingerprint.py -v` — 14 passed
- [x] Broader unit-test suite — no regressions in anything I touched
- [ ] Manual smoke against a real S3 bucket — configure a connector, run
sync twice, expect the second sync to log `bypassed=N, fetched=0` and no
`GetObject` calls in CloudTrail / bucket access logs
- [ ] Manual smoke with `reindex=1` — confirm the full re-download path
still works
### Type of change
- [x] New Feature (non-breaking change which adds functionality)
---------
Co-authored-by: Yingfeng <yingfeng.zhang@gmail.com>
2026-05-09 05:03:56 -07:00
def list_keys ( self ) - > Iterator [ KeyRecord ] :
""" Enumerate the full bucket keyspace with per-object fingerprints.
Cheap path : relies on list_objects_v2 which returns ETag in the listing ,
so no GetObject call is needed . Caches each object ' s metadata so a
subsequent get_value ( key ) call can rebuild the Document without a second
round - trip to S3 .
"""
if self . s3_client is None :
raise ConnectorMissingCredentialError ( " Blob storage " )
all_objects , filename_counts = self . _collect_blob_objects (
start = datetime ( 1970 , 1 , 1 , tzinfo = timezone . utc ) ,
end = datetime . now ( timezone . utc ) ,
)
self . _filename_counts = filename_counts
self . _listing_cache = { }
for obj in all_objects :
doc_id = f " { self . bucket_type } : { self . bucket_name } : { obj [ ' Key ' ] } "
self . _listing_cache [ doc_id ] = obj
yield KeyRecord (
key = doc_id ,
fingerprint = _normalize_etag ( obj . get ( " ETag " ) ) ,
)
def get_value ( self , key : str ) - > Document :
""" Materialize the Document for a key previously yielded by list_keys().
Must be called within the same list_keys ( ) pass that produced the key ,
since the metadata cache lives on the connector instance and is reset
each list_keys ( ) call .
"""
obj = self . _listing_cache . get ( key )
if obj is None :
2026-07-03 12:53:39 +08:00
raise KeyError ( f " get_value( { key !r} ) called before list_keys() yielded the key, or after a subsequent list_keys() reset the cache " )
feat(connectors): ETag-based bypass for incremental S3 ingestion (#14628) (#14677)
### What problem does this PR solve?
S3-family connector syncs currently re-download every in-window object
just so we can compute `xxhash128(blob)` and compare against
`Document.content_hash`. Anything that bumps `LastModified` without
changing bytes (`aws s3 cp` touches, bucket re-encryption, etc.) pays
full bandwidth and re-parses files that didn't actually change. #14628
covers the broader incremental-ingestion redesign; this PR is the first
slice.
The fix is a pre-listing short-circuit. `BlobStorageConnector` (S3 / R2
/ GCS / OCI / S3-compat) now implements a new `FingerprintConnector`
interface: `list_keys()` paginates `list_objects_v2` and yields
`KeyRecord(key, fingerprint)` where `fingerprint = xxhash128(ETag)`. The
orchestrator joins those against the connector's existing `{doc_id:
content_hash}` map and only calls `get_value(key)` when the fingerprint
differs. Unchanged keys are skipped entirely — no `GetObject`, no
re-parse.
No DDL. xxhash128(ETag) is 32 hex chars and reuses the existing
`Document.content_hash` column per @yingfeng's suggestion; the connector
decides at listing time whether to populate it. Local uploads and
connectors that don't opt in fall through to the existing post-download
`xxhash128(blob)` path with no behavior change.
This is PR-1 of a 4-PR series — full design lives on #14628. Subsequent
PRs extend tier 1 to local FS / WebDAV / Dropbox / Seafile / RDBMS
(PR-2), wire up tier 2 cursor connectors with `SyncLogs.next_checkpoint`
(PR-3), and unify deletion via `KeyRecord(deleted=True)` reconciliation
(PR-4). Holding those back keeps this PR additive and reviewable on its
own.
#### Files touched
- `common/data_source/models.py` — new `KeyRecord`; optional
`fingerprint` on `Document`
- `common/data_source/interfaces.py` — `IncrementalCapability` enum,
`FingerprintConnector` ABC
- `common/data_source/blob_connector.py` — `BlobStorageConnector`
implements `FingerprintConnector`; per-object download factored into
`_build_document_from_obj()` so `_yield_blob_objects`, `list_keys`,
`get_value` all share it
- `rag/svr/sync_data_source.py` —
`_BlobLikeBase._fingerprint_filtered_generator` does the bypass loop;
`_run_task_logic` plumbs `doc.fingerprint` into the upload dict
- `api/db/services/document_service.py` —
`list_id_content_hash_map_by_kb_and_source_type()` helper
- `api/db/services/connector_service.py` + `file_service.py` —
fingerprint flows through `duplicate_and_parse → upload_document` and
lands in `content_hash`
- `test/unit_test/common/test_blob_connector_fingerprint.py` — 14 tests
covering ETag normalization (single-part, multipart, quoted, empty),
`list_keys()` not calling `GetObject`, `get_value()` materializing with
fingerprint, deterministic/stable fingerprints, and the bypass loop
asserting `GetObject` is *not* called on a match
#### Worth flagging for review
Old `_BlobLikeBase._generate` called `poll_source(start, now)` with a
`LastModified` window when `poll_range_start` was set. New code uses
`_fingerprint_filtered_generator` (full bucket listing + fingerprint
compare) outside of explicit `reindex=1`. Strictly better for
unchanged-bucket cases since it skips `GetObject`, but it does mean
every sync now does a full `list_objects_v2` paginate. Should still be
cheap for most buckets — flagging in case anyone has a very large bucket
where the time-window filter was meaningful.
On migration: existing rows have `content_hash = xxhash128(blob)` from
the old code. The first sync after this lands sees ETag-derived
fingerprints that don't match, re-fetches every object once, and writes
the new fingerprint. From the second sync onward the bypass works as
expected. "Slow day one, fast every day after." A `fingerprint_backfill:
trust` opt-out is sketched in the design doc but not in this PR.
#### Test plan
- [x] `uv run ruff check` — clean on all 8 touched files
- [x] `uv run pytest
test/unit_test/common/test_blob_connector_fingerprint.py -v` — 14 passed
- [x] Broader unit-test suite — no regressions in anything I touched
- [ ] Manual smoke against a real S3 bucket — configure a connector, run
sync twice, expect the second sync to log `bypassed=N, fetched=0` and no
`GetObject` calls in CloudTrail / bucket access logs
- [ ] Manual smoke with `reindex=1` — confirm the full re-download path
still works
### Type of change
- [x] New Feature (non-breaking change which adds functionality)
---------
Co-authored-by: Yingfeng <yingfeng.zhang@gmail.com>
2026-05-09 05:03:56 -07:00
doc = self . _build_document_from_obj ( obj , self . _filename_counts )
if doc is None :
raise RuntimeError ( f " Failed to materialize Document for key { key !r} " )
return doc
2026-04-28 15:07:14 +08:00
def _collect_blob_objects (
self ,
start : datetime ,
end : datetime ,
) - > tuple [ list [ dict [ str , Any ] ] , dict [ str , int ] ] :
""" Collect object metadata for files in the requested window. """
if self . s3_client is None :
raise ConnectorMissingCredentialError ( " Blob storage " )
paginator = self . s3_client . get_paginator ( " list_objects_v2 " )
pages = paginator . paginate ( Bucket = self . bucket_name , Prefix = self . prefix )
# Collect all objects first to count filename occurrences
all_objects : list [ dict [ str , Any ] ] = [ ]
for page in pages :
if " Contents " not in page :
continue
for obj in page [ " Contents " ] :
if obj [ " Key " ] . endswith ( " / " ) :
continue
last_modified = obj [ " LastModified " ] . replace ( tzinfo = timezone . utc )
if start < last_modified < = end :
all_objects . append ( obj )
filename_counts : dict [ str , int ] = { }
for obj in all_objects :
file_name = os . path . basename ( obj [ " Key " ] )
filename_counts [ file_name ] = filename_counts . get ( file_name , 0 ) + 1
return all_objects , filename_counts
def _get_semantic_id (
self ,
key : str ,
file_name : str ,
filename_counts : dict [ str , int ] ,
) - > str :
""" Use full relative path only when filenames collide. """
if filename_counts . get ( file_name , 0 ) > 1 :
relative_path = key
if self . prefix and key . startswith ( self . prefix ) :
2026-07-03 12:53:39 +08:00
relative_path = key [ len ( self . prefix ) : ]
2026-04-28 15:07:14 +08:00
return relative_path . replace ( " / " , " / " ) if relative_path else file_name
return file_name
def retrieve_all_slim_docs_perm_sync (
self ,
callback : Any = None ,
) - > GenerateSlimDocumentOutput :
""" Return a full current snapshot of blob object IDs without downloading content. """
del callback
all_objects , _ = self . _collect_blob_objects (
start = datetime ( 1970 , 1 , 1 , tzinfo = timezone . utc ) ,
end = datetime . now ( timezone . utc ) ,
)
batch : list [ SlimDocument ] = [ ]
for obj in all_objects :
2026-07-03 12:53:39 +08:00
batch . append ( SlimDocument ( id = f " { self . bucket_type } : { self . bucket_name } : { obj [ ' Key ' ] } " ) )
2026-04-28 15:07:14 +08:00
if len ( batch ) == self . batch_size :
yield batch
batch = [ ]
2025-11-03 19:59:18 +08:00
if batch :
yield batch
def load_from_state ( self ) - > GenerateDocumentsOutput :
""" Load documents from state """
logging . debug ( " Loading blob objects " )
return self . _yield_blob_objects (
start = datetime ( 1970 , 1 , 1 , tzinfo = timezone . utc ) ,
end = datetime . now ( timezone . utc ) ,
)
2026-07-03 12:53:39 +08:00
def poll_source ( self , start : SecondsSinceUnixEpoch , end : SecondsSinceUnixEpoch ) - > GenerateDocumentsOutput :
2025-11-03 19:59:18 +08:00
""" Poll source to get documents """
if self . s3_client is None :
raise ConnectorMissingCredentialError ( " Blob storage " )
start_datetime = datetime . fromtimestamp ( start , tz = timezone . utc )
end_datetime = datetime . fromtimestamp ( end , tz = timezone . utc )
for batch in self . _yield_blob_objects ( start_datetime , end_datetime ) :
yield batch
def validate_connector_settings ( self ) - > None :
""" Validate connector settings """
if self . s3_client is None :
2026-07-03 12:53:39 +08:00
raise ConnectorMissingCredentialError ( " Blob storage credentials not loaded. " )
2025-11-03 19:59:18 +08:00
if not self . bucket_name :
2026-07-03 12:53:39 +08:00
raise ConnectorValidationError ( " No bucket name was provided in connector settings. " )
2025-11-03 19:59:18 +08:00
try :
# Lightweight validation step
2026-07-03 12:53:39 +08:00
self . s3_client . list_objects_v2 ( Bucket = self . bucket_name , Prefix = self . prefix , MaxKeys = 1 )
2025-11-03 19:59:18 +08:00
except Exception as e :
2026-07-03 12:53:39 +08:00
error_code = getattr ( e , " response " , { } ) . get ( " Error " , { } ) . get ( " Code " , " " )
status_code = getattr ( e , " response " , { } ) . get ( " ResponseMetadata " , { } ) . get ( " HTTPStatusCode " )
2025-11-03 19:59:18 +08:00
# Common S3 error scenarios
if error_code in [
" AccessDenied " ,
" InvalidAccessKeyId " ,
" SignatureDoesNotMatch " ,
] :
if status_code == 403 or error_code == " AccessDenied " :
2026-07-03 12:53:39 +08:00
raise InsufficientPermissionsError ( f " Insufficient permissions to list objects in bucket ' { self . bucket_name } ' . Please check your bucket policy and/or IAM policy. " )
2025-11-03 19:59:18 +08:00
if status_code == 401 or error_code == " SignatureDoesNotMatch " :
2026-07-03 12:53:39 +08:00
raise CredentialExpiredError ( " Provided blob storage credentials appear invalid or expired. " )
2025-11-03 19:59:18 +08:00
2026-07-03 12:53:39 +08:00
raise CredentialExpiredError ( f " Credential issue encountered ( { error_code } ). " )
2025-11-03 19:59:18 +08:00
if error_code == " NoSuchBucket " or status_code == 404 :
2026-07-03 12:53:39 +08:00
raise ConnectorValidationError ( f " Bucket ' { self . bucket_name } ' does not exist or cannot be found. " )
2025-11-03 19:59:18 +08:00
2026-07-03 12:53:39 +08:00
raise ConnectorValidationError ( f " Unexpected S3 client error (code= { error_code } , status= { status_code } ): { e } " )
2025-11-03 19:59:18 +08:00
if __name__ == " __main__ " :
# Example usage
credentials_dict = {
" aws_access_key_id " : os . environ . get ( " AWS_ACCESS_KEY_ID " ) ,
" aws_secret_access_key " : os . environ . get ( " AWS_SECRET_ACCESS_KEY " ) ,
}
# Initialize connector
connector = BlobStorageConnector (
bucket_type = os . environ . get ( " BUCKET_TYPE " ) or " s3 " ,
bucket_name = os . environ . get ( " BUCKET_NAME " ) or " yyboombucket " ,
prefix = " " ,
)
try :
connector . load_credentials ( credentials_dict )
document_batch_generator = connector . load_from_state ( )
for document_batch in document_batch_generator :
print ( " First batch of documents: " )
for doc in document_batch :
print ( f " Document ID: { doc . id } " )
print ( f " Semantic Identifier: { doc . semantic_identifier } " )
print ( f " Source: { doc . source } " )
print ( f " Updated At: { doc . doc_updated_at } " )
print ( " --- " )
break
except ConnectorMissingCredentialError as e :
print ( f " Error: { e } " )
except Exception as e :
2025-12-22 19:06:01 +08:00
print ( f " An unexpected error occurred: { e } " )