Files
ragflow/internal/tokenizer/tokenizer_test.go
Jack e997fd655a fix(tokenizer): load cl100k BPE table from disk instead of failing silently offline (#17712)
## Summary

RAGFlow's Go tokenizer silently returned **0 tokens for every string**
whenever the `cl100k_base` BPE table could not be loaded — which is the
normal case for an offline/air-gapped Go server. This PR makes the
loader resolve the table from disk (where RAGFlow actually ships it) and
fail loudly when it is genuinely missing.

## Root cause

`tiktoken-go`'s stock loader downloads the encoding table over HTTP and
caches it under `TIKTOKEN_CACHE_DIR`. That does not work for RAGFlow:

- `TIKTOKEN_CACHE_DIR` is exported **only inside the Python process**
(`common/token_utils.py`). `docker/entrypoint.sh` launches the Go binary
(`bin/ragflow_server`) from a shell, so the Go process never inherits
the variable.
- The Dockerfile *does* ship the table (under its sha1 name in the
working directory), but nothing told the Go side to look there.
- Reaching `openaipublic.blob.core.windows.net` at runtime is not an
option for air-gapped installs, and is unreliable where that host is
blocked.

The failure was **silent**: `NumTokensFromString` returns `0` when the
encoder fails to build, and a `sync.Once` memoizes that error for the
process lifetime. Every token count became `0`, so chunk merging never
crossed its token budget and an entire document collapsed into a single
chunk. Python has no such failure mode because its encoder is built at
import time (a missing table aborts startup instead of degrading).

## Fix

Register a local-only `BpeLoader` via `tiktoken.SetBpeLoader`
(`internal/tokenizer/bpe_loader.go`) that resolves the table from disk
**only**, in priority order:

1. `TIKTOKEN_CACHE_DIR` / `DATA_GYM_CACHE_DIR` (honored so operators who
already configured one keep working).
2. The working directory, the executable's directory, and all of their
ancestors — matching the Dockerfile layout (table under its sha1 name in
the install root).
3. A `ragflow_deps/<basename>` checkout produced by
`ragflow_deps/download_deps.py`.

It **never performs network I/O**. When nothing is found it returns an
error listing every path it tried (pointing at `download_deps.py` or
`TIKTOKEN_CACHE_DIR`), so a genuinely missing table fails loudly instead
of degrading to zero.

## Test plan

- `internal/tokenizer/bpe_loader_test.go` (unit tier, runs under `bash
build.sh --test ./internal/tokenizer/...`):
- Loader reads from `TIKTOKEN_CACHE_DIR`, `DATA_GYM_CACHE_DIR`, the
sha1-named file in the working dir, and the bundled `ragflow_deps/`
name.
  - Explicit cache dir wins over the bundled vocab.
  - A malformed table is reported as an error rather than skipped.
- A genuinely missing table reports the candidates it tried (no network
attempt).
- `NumTokensFromString` matches Python-derived anchors (`""`→0,
`"hello"`→1, `"hello world"`→2, `"hello, world!"`→4, `"世界"`→3, `"Hello
世界 🌍"`→8, `"RAGFlow"`→3).

## Notes

- `.github/workflows/tests.yml` currently excludes `internal/tokenizer`
from `go test`, so these tests do not run in CI. The tokenizer fix is
exercised in CI indirectly via the chunker package once a
token-count-sensitive parity case lands (tracked separately). Consider
including `internal/tokenizer` in CI as a follow-up.
- Supported deployments already ship the table (`download_deps.py` →
`ragflow_deps/cl100k_base.tiktoken`; Dockerfile → `<sha1>` in cwd), so
no `ENV` change is required for the fix to take effect. Setting `ENV
TIKTOKEN_CACHE_DIR` in the Dockerfile remains a cheap
belt-and-suspenders hardening that can be done separately.

🤖 Generated with [CodeBuddy Code](https://cnb.cool/codebuddy)

---------

Co-authored-by: CodeBuddy <noreply@codebuddy.ai>
Co-authored-by: CodeBuddy Code <noreply@cnb.cool>
Co-authored-by: CodeBuddy <noreply@tencent.com>
2026-08-03 19:03:08 +08:00

346 lines
9.5 KiB
Go

//
// Copyright 2026 The InfiniFlow Authors. All Rights Reserved.
//
// Licensed under the Apache License, Version 2.0 (the "License");
// you may not use this file except in compliance with the License.
// You may obtain a copy of the License at
//
// http://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing, software
// distributed under the License is distributed on an "AS IS" BASIS,
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
// See the License for the specific language governing permissions and
// limitations under the License.
//go:build manual
package tokenizer
import (
"strings"
"testing"
"time"
)
type languageDifferentiator struct {
input string
english string
dutch string
}
// saveEngineType saves the current engineTypeProvider and returns a function
// to restore it. Use this when a test modifies the engine type to avoid
// leaking global state between tests.
func saveEngineType() func() {
original := engineType
return func() { engineType = original }
}
// ---------------------------------------------------------------------------
// NumTokensFromString tests
// ---------------------------------------------------------------------------
func TestNumTokensFromString_Empty(t *testing.T) {
if got := NumTokensFromString(""); got != 0 {
t.Errorf("expected 0 for empty string, got %d", got)
}
}
func TestNumTokensFromString_Positive(t *testing.T) {
for _, s := range []string{"hello world", "你好世界"} {
if got := NumTokensFromString(s); got <= 0 {
t.Errorf("NumTokensFromString(%q) = %d, want >0", s, got)
}
}
}
func TestNumTokensFromString_VariedInputs(t *testing.T) {
tests := []struct {
name string
input string
}{
{"ascii letters", "hello world"},
{"chinese characters", "你好世界"},
{"japanese characters", "こんにちは世界"},
{"korean characters", "안녕하세요세계"},
{"emoji", "👋 hello 🌍"},
{"numbers only", "1234567890"},
{"special chars", "a+b=c; d!=e"},
{"newlines and tabs", "line1\nline2\tindented"},
{"mixed content", "RAGFlow 是一款 开源的 RAG (Retrieval-Augmented Generation) 引擎"},
}
for _, tt := range tests {
t.Run(tt.name, func(t *testing.T) {
got := NumTokensFromString(tt.input)
if got <= 0 {
t.Errorf("NumTokensFromString(%q) = %d, want >0", tt.input, got)
}
})
}
}
func TestNumTokensFromString_Consistency(t *testing.T) {
inputs := []string{"hello world", "你好世界", "a+b=c; d!=e"}
for _, s := range inputs {
first := NumTokensFromString(s)
second := NumTokensFromString(s)
if first != second {
t.Errorf("NumTokensFromString(%q) is not consistent: %d vs %d", s, first, second)
}
}
}
func TestNumTokensFromString_LongString(t *testing.T) {
long := strings.Repeat("the quick brown fox jumps over the lazy dog. ", 200)
got := NumTokensFromString(long)
if got <= 0 {
t.Errorf("NumTokensFromString(long_string) = %d, want >0", got)
}
}
func TestNumTokensFromString_WhitespaceOnly(t *testing.T) {
for _, s := range []string{" ", "\t", "\n", " "} {
got := NumTokensFromString(s)
// Whitespace strings should still produce tokens in BPE encoding
if got == 0 {
t.Logf("NumTokensFromString(%q) = %d", s, got)
}
}
}
// ---------------------------------------------------------------------------
// RegisterEngineType tests
// ---------------------------------------------------------------------------
func TestRegisterEngineType_Basic(t *testing.T) {
restore := saveEngineType()
defer restore()
SetEngineType("infinity")
if got := engineType; got != "infinity" {
t.Errorf("expected 'infinity', got %q", got)
}
}
func TestRegisterEngineType_Overwrite(t *testing.T) {
restore := saveEngineType()
defer restore()
SetEngineType("first")
SetEngineType("second")
if got := engineType; got != "second" {
t.Errorf("expected 'second', got %q", got)
}
}
// ---------------------------------------------------------------------------
// Tokenize tests
// ---------------------------------------------------------------------------
func TestTokenize_InfinityEngine(t *testing.T) {
restore := saveEngineType()
defer restore()
SetEngineType("infinity")
inputs := []string{"hello world", "你好 世界", "", "a single word"}
for _, input := range inputs {
got, err := Tokenize(input)
if err != nil {
t.Errorf("Tokenize(%q) unexpected error: %v", input, err)
}
if got != input {
t.Errorf("Tokenize(%q) = %q, want %q", input, got, input)
}
}
}
func TestTokenize_PoolNotInitialized(t *testing.T) {
restore := saveEngineType()
defer restore()
// Ensure engine type is not "infinity" so we hit the pool path
SetEngineType("")
_, err := Tokenize("hello world")
if err == nil {
t.Error("expected error when pool is not initialized, got nil")
}
}
// ---------------------------------------------------------------------------
// FineGrainedTokenize tests
// ---------------------------------------------------------------------------
func TestFineGrainedTokenize_InfinityEngine(t *testing.T) {
restore := saveEngineType()
defer restore()
SetEngineType("infinity")
inputs := []string{"hello world", "测试 分词", ""}
for _, input := range inputs {
got, err := FineGrainedTokenize(input)
if err != nil {
t.Errorf("FineGrainedTokenize(%q) unexpected error: %v", input, err)
}
if got != input {
t.Errorf("FineGrainedTokenize(%q) = %q, want %q", input, got, input)
}
}
}
func TestFineGrainedTokenize_PoolNotInitialized(t *testing.T) {
restore := saveEngineType()
defer restore()
SetEngineType("")
_, err := FineGrainedTokenize("hello world")
if err == nil {
t.Error("expected error when pool is not initialized, got nil")
}
}
// ---------------------------------------------------------------------------
// Error-path tests for functions that require the pool
// ---------------------------------------------------------------------------
func TestTokenizeWithPosition_PoolNotInitialized(t *testing.T) {
_, err := TokenizeWithPosition("hello world")
if err == nil {
t.Error("expected error when pool is not initialized, got nil")
}
}
func TestAnalyze_PoolNotInitialized(t *testing.T) {
_, err := Analyze("hello world")
if err == nil {
t.Error("expected error when pool is not initialized, got nil")
}
}
func TestGetTermFreq_PoolNotInitialized(t *testing.T) {
got := GetTermFreq("hello")
if got != 0 {
t.Errorf("expected 0 when pool is not initialized, got %d", got)
}
}
func TestGetTermTag_PoolNotInitialized(t *testing.T) {
got := GetTermTag("hello")
if got != "" {
t.Errorf("expected empty string when pool is not initialized, got %q", got)
}
}
func TestTokenize_DefaultLanguageResetsAnalyzerState(t *testing.T) {
restore := saveEngineType()
defer restore()
SetEngineType("")
if err := Init(&PoolConfig{
DictPath: "",
MinSize: 1,
MaxSize: 1,
IdleTimeout: 5 * time.Second,
AcquireTimeout: 5 * time.Second,
}); err != nil {
t.Fatalf("Failed to initialize pool: %v", err)
}
defer Close()
sample := findEnglishDutchDifferentiator(t)
dutchGot, err := New("Dutch").Tokenize(sample.input)
if err != nil {
t.Fatalf("Tokenize(Dutch, %q) unexpected error: %v", sample.input, err)
}
if dutchGot != sample.dutch {
t.Fatalf("Tokenize(Dutch, %q) = %q, want %q", sample.input, dutchGot, sample.dutch)
}
defaultGot, err := Tokenize(sample.input)
if err != nil {
t.Fatalf("Tokenize(default, %q) unexpected error: %v", sample.input, err)
}
if defaultGot != sample.english {
t.Fatalf("Tokenize(default, %q) = %q, want explicit English result %q", sample.input, defaultGot, sample.english)
}
if defaultGot == dutchGot {
t.Fatalf("Tokenize(default, %q) unexpectedly inherited Dutch analyzer state: %q", sample.input, defaultGot)
}
}
func findEnglishDutchDifferentiator(t *testing.T) languageDifferentiator {
t.Helper()
candidates := []string{
"running",
"jumps",
"ponies",
"studies",
"wolves",
"relational",
"conditionally",
}
for _, input := range candidates {
english, err := New("English").Tokenize(input)
if err != nil {
t.Fatalf("Tokenize(English, %q) unexpected error: %v", input, err)
}
dutch, err := New("Dutch").Tokenize(input)
if err != nil {
t.Fatalf("Tokenize(Dutch, %q) unexpected error: %v", input, err)
}
if english != dutch {
return languageDifferentiator{
input: input,
english: english,
dutch: dutch,
}
}
}
t.Skip("no differentiating tokenizer sample found for English vs Dutch")
return languageDifferentiator{}
}
// ---------------------------------------------------------------------------
// Global state tests
// ---------------------------------------------------------------------------
func TestGetPoolStats_Nil(t *testing.T) {
// Note: globalPool is nil by default in unit tests (pool not initialized)
stats := GetPoolStats()
if stats == nil {
t.Fatal("GetPoolStats returned nil")
}
init, ok := stats["initialized"]
if !ok {
t.Fatal("missing 'initialized' key")
}
if init.(bool) {
t.Error("expected initialized=false when pool is nil")
}
}
func TestIsInitialized_Default(t *testing.T) {
if IsInitialized() {
t.Error("expected IsInitialized() = false when pool is not initialized")
}
}
func TestClose_Nil(t *testing.T) {
// Close should be safe to call with nil globalPool
Close() // no panic = pass
}
func TestClose_NilGlobalPool(t *testing.T) {
// Call Close directly after ensuring globalPool is nil
// (concurrent test may have initialized it, so handle gracefully)
defer func() {
if r := recover(); r != nil {
t.Errorf("Close() panicked: %v", r)
}
}()
Close()
}