forkjo/modules/indexer/code/elastic_search.go

525 lines
15 KiB
Go
Raw Normal View History

// Copyright 2020 The Gitea Authors. All rights reserved.
// SPDX-License-Identifier: MIT
package code
import (
"bufio"
"context"
"errors"
"fmt"
"io"
"net"
"strconv"
"strings"
"sync"
"time"
repo_model "code.gitea.io/gitea/models/repo"
"code.gitea.io/gitea/modules/analyze"
"code.gitea.io/gitea/modules/charset"
"code.gitea.io/gitea/modules/git"
"code.gitea.io/gitea/modules/graceful"
"code.gitea.io/gitea/modules/json"
"code.gitea.io/gitea/modules/log"
"code.gitea.io/gitea/modules/setting"
"code.gitea.io/gitea/modules/timeutil"
"code.gitea.io/gitea/modules/typesniffer"
"github.com/go-enry/go-enry/v2"
"github.com/olivere/elastic/v7"
)
const (
esRepoIndexerLatestVersion = 1
// multi-match-types, currently only 2 types are used
// Reference: https://www.elastic.co/guide/en/elasticsearch/reference/7.0/query-dsl-multi-match-query.html#multi-match-types
esMultiMatchTypeBestFields = "best_fields"
esMultiMatchTypePhrasePrefix = "phrase_prefix"
)
var _ Indexer = &ElasticSearchIndexer{}
// ElasticSearchIndexer implements Indexer interface
type ElasticSearchIndexer struct {
Rewrite queue (#24505) # ⚠️ Breaking Many deprecated queue config options are removed (actually, they should have been removed in 1.18/1.19). If you see the fatal message when starting Gitea: "Please update your app.ini to remove deprecated config options", please follow the error messages to remove these options from your app.ini. Example: ``` 2023/05/06 19:39:22 [E] Removed queue option: `[indexer].ISSUE_INDEXER_QUEUE_TYPE`. Use new options in `[queue.issue_indexer]` 2023/05/06 19:39:22 [E] Removed queue option: `[indexer].UPDATE_BUFFER_LEN`. Use new options in `[queue.issue_indexer]` 2023/05/06 19:39:22 [F] Please update your app.ini to remove deprecated config options ``` Many options in `[queue]` are are dropped, including: `WRAP_IF_NECESSARY`, `MAX_ATTEMPTS`, `TIMEOUT`, `WORKERS`, `BLOCK_TIMEOUT`, `BOOST_TIMEOUT`, `BOOST_WORKERS`, they can be removed from app.ini. # The problem The old queue package has some legacy problems: * complexity: I doubt few people could tell how it works. * maintainability: Too many channels and mutex/cond are mixed together, too many different structs/interfaces depends each other. * stability: due to the complexity & maintainability, sometimes there are strange bugs and difficult to debug, and some code doesn't have test (indeed some code is difficult to test because a lot of things are mixed together). * general applicability: although it is called "queue", its behavior is not a well-known queue. * scalability: it doesn't seem easy to make it work with a cluster without breaking its behaviors. It came from some very old code to "avoid breaking", however, its technical debt is too heavy now. It's a good time to introduce a better "queue" package. # The new queue package It keeps using old config and concept as much as possible. * It only contains two major kinds of concepts: * The "base queue": channel, levelqueue, redis * They have the same abstraction, the same interface, and they are tested by the same testing code. * The "WokerPoolQueue", it uses the "base queue" to provide "worker pool" function, calls the "handler" to process the data in the base queue. * The new code doesn't do "PushBack" * Think about a queue with many workers, the "PushBack" can't guarantee the order for re-queued unhandled items, so in new code it just does "normal push" * The new code doesn't do "pause/resume" * The "pause/resume" was designed to handle some handler's failure: eg: document indexer (elasticsearch) is down * If a queue is paused for long time, either the producers blocks or the new items are dropped. * The new code doesn't do such "pause/resume" trick, it's not a common queue's behavior and it doesn't help much. * If there are unhandled items, the "push" function just blocks for a few seconds and then re-queue them and retry. * The new code doesn't do "worker booster" * Gitea's queue's handlers are light functions, the cost is only the go-routine, so it doesn't make sense to "boost" them. * The new code only use "max worker number" to limit the concurrent workers. * The new "Push" never blocks forever * Instead of creating more and more blocking goroutines, return an error is more friendly to the server and to the end user. There are more details in code comments: eg: the "Flush" problem, the strange "code.index" hanging problem, the "immediate" queue problem. Almost ready for review. TODO: * [x] add some necessary comments during review * [x] add some more tests if necessary * [x] update documents and config options * [x] test max worker / active worker * [x] re-run the CI tasks to see whether any test is flaky * [x] improve the `handleOldLengthConfiguration` to provide more friendly messages * [x] fine tune default config values (eg: length?) ## Code coverage: ![image](https://user-images.githubusercontent.com/2114189/236620635-55576955-f95d-4810-b12f-879026a3afdf.png)
2023-05-08 13:49:59 +02:00
client *elastic.Client
indexerAliasName string
available bool
stopTimer chan struct{}
lock sync.RWMutex
}
type elasticLogger struct {
log.Logger
}
func (l elasticLogger) Printf(format string, args ...interface{}) {
_ = l.Logger.Log(2, l.Logger.GetLevel(), format, args...)
}
// NewElasticSearchIndexer creates a new elasticsearch indexer
func NewElasticSearchIndexer(url, indexerName string) (*ElasticSearchIndexer, bool, error) {
opts := []elastic.ClientOptionFunc{
elastic.SetURL(url),
elastic.SetSniff(false),
elastic.SetHealthcheckInterval(10 * time.Second),
elastic.SetGzip(false),
}
logger := elasticLogger{log.GetLogger(log.DEFAULT)}
if logger.GetLevel() == log.TRACE || logger.GetLevel() == log.DEBUG {
opts = append(opts, elastic.SetTraceLog(logger))
} else if logger.GetLevel() == log.ERROR || logger.GetLevel() == log.CRITICAL || logger.GetLevel() == log.FATAL {
opts = append(opts, elastic.SetErrorLog(logger))
} else if logger.GetLevel() == log.INFO || logger.GetLevel() == log.WARN {
opts = append(opts, elastic.SetInfoLog(logger))
}
client, err := elastic.NewClient(opts...)
if err != nil {
return nil, false, err
}
indexer := &ElasticSearchIndexer{
client: client,
indexerAliasName: indexerName,
available: true,
stopTimer: make(chan struct{}),
}
ticker := time.NewTicker(10 * time.Second)
go func() {
for {
select {
case <-ticker.C:
indexer.checkAvailability()
case <-indexer.stopTimer:
ticker.Stop()
return
}
}
}()
exists, err := indexer.init()
if err != nil {
indexer.Close()
return nil, false, err
}
return indexer, !exists, err
}
const (
defaultMapping = `{
"mappings": {
"properties": {
"repo_id": {
"type": "long",
"index": true
},
"content": {
"type": "text",
"term_vector": "with_positions_offsets",
"index": true
},
"commit_id": {
"type": "keyword",
"index": true
},
"language": {
"type": "keyword",
"index": true
},
"updated_at": {
"type": "long",
"index": true
}
}
}
}`
)
func (b *ElasticSearchIndexer) realIndexerName() string {
return fmt.Sprintf("%s.v%d", b.indexerAliasName, esRepoIndexerLatestVersion)
}
// Init will initialize the indexer
func (b *ElasticSearchIndexer) init() (bool, error) {
ctx := graceful.GetManager().HammerContext()
exists, err := b.client.IndexExists(b.realIndexerName()).Do(ctx)
if err != nil {
return false, b.checkError(err)
}
if !exists {
mapping := defaultMapping
createIndex, err := b.client.CreateIndex(b.realIndexerName()).BodyString(mapping).Do(ctx)
if err != nil {
return false, b.checkError(err)
}
if !createIndex.Acknowledged {
return false, fmt.Errorf("create index %s with %s failed", b.realIndexerName(), mapping)
}
}
// check version
r, err := b.client.Aliases().Do(ctx)
if err != nil {
return false, b.checkError(err)
}
realIndexerNames := r.IndicesByAlias(b.indexerAliasName)
if len(realIndexerNames) < 1 {
res, err := b.client.Alias().
Add(b.realIndexerName(), b.indexerAliasName).
Do(ctx)
if err != nil {
return false, b.checkError(err)
}
if !res.Acknowledged {
return false, fmt.Errorf("create alias %s to index %s failed", b.indexerAliasName, b.realIndexerName())
}
} else if len(realIndexerNames) >= 1 && realIndexerNames[0] < b.realIndexerName() {
log.Warn("Found older gitea indexer named %s, but we will create a new one %s and keep the old NOT DELETED. You can delete the old version after the upgrade succeed.",
realIndexerNames[0], b.realIndexerName())
res, err := b.client.Alias().
Remove(realIndexerNames[0], b.indexerAliasName).
Add(b.realIndexerName(), b.indexerAliasName).
Do(ctx)
if err != nil {
return false, b.checkError(err)
}
if !res.Acknowledged {
return false, fmt.Errorf("change alias %s to index %s failed", b.indexerAliasName, b.realIndexerName())
}
}
return exists, nil
}
// Ping checks if elastic is available
func (b *ElasticSearchIndexer) Ping() bool {
b.lock.RLock()
defer b.lock.RUnlock()
return b.available
}
func (b *ElasticSearchIndexer) addUpdate(ctx context.Context, batchWriter git.WriteCloserError, batchReader *bufio.Reader, sha string, update fileUpdate, repo *repo_model.Repository) ([]elastic.BulkableRequest, error) {
// Ignore vendored files in code search
if setting.Indexer.ExcludeVendored && analyze.IsVendor(update.Filename) {
return nil, nil
}
size := update.Size
var err error
if !update.Sized {
var stdout string
stdout, _, err = git.NewCommand(ctx, "cat-file", "-s").AddDynamicArguments(update.BlobSha).RunStdString(&git.RunOpts{Dir: repo.RepoPath()})
if err != nil {
return nil, err
}
if size, err = strconv.ParseInt(strings.TrimSpace(stdout), 10, 64); err != nil {
return nil, fmt.Errorf("misformatted git cat-file output: %w", err)
}
}
if size > setting.Indexer.MaxIndexerFileSize {
return []elastic.BulkableRequest{b.addDelete(update.Filename, repo)}, nil
}
if _, err := batchWriter.Write([]byte(update.BlobSha + "\n")); err != nil {
return nil, err
}
_, _, size, err = git.ReadBatchLine(batchReader)
if err != nil {
return nil, err
}
fileContents, err := io.ReadAll(io.LimitReader(batchReader, size))
if err != nil {
return nil, err
} else if !typesniffer.DetectContentType(fileContents).IsText() {
// FIXME: UTF-16 files will probably fail here
return nil, nil
}
if _, err = batchReader.Discard(1); err != nil {
return nil, err
}
id := filenameIndexerID(repo.ID, update.Filename)
return []elastic.BulkableRequest{
elastic.NewBulkIndexRequest().
Index(b.indexerAliasName).
Id(id).
Doc(map[string]interface{}{
"repo_id": repo.ID,
"content": string(charset.ToUTF8DropErrors(fileContents)),
"commit_id": sha,
"language": analyze.GetCodeLanguage(update.Filename, fileContents),
"updated_at": timeutil.TimeStampNow(),
}),
}, nil
}
func (b *ElasticSearchIndexer) addDelete(filename string, repo *repo_model.Repository) elastic.BulkableRequest {
id := filenameIndexerID(repo.ID, filename)
return elastic.NewBulkDeleteRequest().
Index(b.indexerAliasName).
Id(id)
}
// Index will save the index data
func (b *ElasticSearchIndexer) Index(ctx context.Context, repo *repo_model.Repository, sha string, changes *repoChanges) error {
reqs := make([]elastic.BulkableRequest, 0)
if len(changes.Updates) > 0 {
// Now because of some insanity with git cat-file not immediately failing if not run in a valid git directory we need to run git rev-parse first!
if err := git.EnsureValidGitRepository(ctx, repo.RepoPath()); err != nil {
log.Error("Unable to open git repo: %s for %-v: %v", repo.RepoPath(), repo, err)
return err
}
batchWriter, batchReader, cancel := git.CatFileBatch(ctx, repo.RepoPath())
defer cancel()
for _, update := range changes.Updates {
updateReqs, err := b.addUpdate(ctx, batchWriter, batchReader, sha, update, repo)
if err != nil {
return err
}
if len(updateReqs) > 0 {
reqs = append(reqs, updateReqs...)
}
}
cancel()
}
for _, filename := range changes.RemovedFilenames {
reqs = append(reqs, b.addDelete(filename, repo))
}
if len(reqs) > 0 {
_, err := b.client.Bulk().
Index(b.indexerAliasName).
Add(reqs...).
Do(ctx)
return b.checkError(err)
}
return nil
}
// Delete deletes indexes by ids
func (b *ElasticSearchIndexer) Delete(repoID int64) error {
_, err := b.client.DeleteByQuery(b.indexerAliasName).
Query(elastic.NewTermsQuery("repo_id", repoID)).
Do(graceful.GetManager().HammerContext())
return b.checkError(err)
}
// indexPos find words positions for start and the following end on content. It will
// return the beginning position of the first start and the ending position of the
// first end following the start string.
// If not found any of the positions, it will return -1, -1.
func indexPos(content, start, end string) (int, int) {
startIdx := strings.Index(content, start)
if startIdx < 0 {
return -1, -1
}
endIdx := strings.Index(content[startIdx+len(start):], end)
if endIdx < 0 {
return -1, -1
}
return startIdx, startIdx + len(start) + endIdx + len(end)
}
func convertResult(searchResult *elastic.SearchResult, kw string, pageSize int) (int64, []*SearchResult, []*SearchResultLanguages, error) {
hits := make([]*SearchResult, 0, pageSize)
for _, hit := range searchResult.Hits.Hits {
// FIXME: There is no way to get the position the keyword on the content currently on the same request.
// So we get it from content, this may made the query slower. See
// https://discuss.elastic.co/t/fetching-position-of-keyword-in-matched-document/94291
var startIndex, endIndex int
c, ok := hit.Highlight["content"]
if ok && len(c) > 0 {
// FIXME: Since the highlighting content will include <em> and </em> for the keywords,
// now we should find the positions. But how to avoid html content which contains the
// <em> and </em> tags? If elastic search has handled that?
startIndex, endIndex = indexPos(c[0], "<em>", "</em>")
if startIndex == -1 {
panic(fmt.Sprintf("1===%s,,,%#v,,,%s", kw, hit.Highlight, c[0]))
}
} else {
panic(fmt.Sprintf("2===%#v", hit.Highlight))
}
repoID, fileName := parseIndexerID(hit.Id)
res := make(map[string]interface{})
if err := json.Unmarshal(hit.Source, &res); err != nil {
return 0, nil, nil, err
}
language := res["language"].(string)
hits = append(hits, &SearchResult{
RepoID: repoID,
Filename: fileName,
CommitID: res["commit_id"].(string),
Content: res["content"].(string),
UpdatedUnix: timeutil.TimeStamp(res["updated_at"].(float64)),
Language: language,
StartIndex: startIndex,
EndIndex: endIndex - 9, // remove the length <em></em> since we give Content the original data
Color: enry.GetColor(language),
})
}
return searchResult.TotalHits(), hits, extractAggs(searchResult), nil
}
func extractAggs(searchResult *elastic.SearchResult) []*SearchResultLanguages {
var searchResultLanguages []*SearchResultLanguages
agg, found := searchResult.Aggregations.Terms("language")
if found {
searchResultLanguages = make([]*SearchResultLanguages, 0, 10)
for _, bucket := range agg.Buckets {
searchResultLanguages = append(searchResultLanguages, &SearchResultLanguages{
Language: bucket.Key.(string),
Color: enry.GetColor(bucket.Key.(string)),
Count: int(bucket.DocCount),
})
}
}
return searchResultLanguages
}
// Search searches for codes and language stats by given conditions.
func (b *ElasticSearchIndexer) Search(ctx context.Context, repoIDs []int64, language, keyword string, page, pageSize int, isMatch bool) (int64, []*SearchResult, []*SearchResultLanguages, error) {
searchType := esMultiMatchTypeBestFields
if isMatch {
searchType = esMultiMatchTypePhrasePrefix
}
kwQuery := elastic.NewMultiMatchQuery(keyword, "content").Type(searchType)
query := elastic.NewBoolQuery()
query = query.Must(kwQuery)
if len(repoIDs) > 0 {
repoStrs := make([]interface{}, 0, len(repoIDs))
for _, repoID := range repoIDs {
repoStrs = append(repoStrs, repoID)
}
repoQuery := elastic.NewTermsQuery("repo_id", repoStrs...)
query = query.Must(repoQuery)
}
var (
start int
kw = "<em>" + keyword + "</em>"
aggregation = elastic.NewTermsAggregation().Field("language").Size(10).OrderByCountDesc()
)
if page > 0 {
start = (page - 1) * pageSize
}
if len(language) == 0 {
searchResult, err := b.client.Search().
Index(b.indexerAliasName).
Aggregation("language", aggregation).
Query(query).
Highlight(
elastic.NewHighlight().
Field("content").
NumOfFragments(0). // return all highting content on fragments
HighlighterType("fvh"),
).
Sort("repo_id", true).
From(start).Size(pageSize).
Do(ctx)
if err != nil {
return 0, nil, nil, b.checkError(err)
}
return convertResult(searchResult, kw, pageSize)
}
langQuery := elastic.NewMatchQuery("language", language)
countResult, err := b.client.Search().
Index(b.indexerAliasName).
Aggregation("language", aggregation).
Query(query).
Size(0). // We only needs stats information
Do(ctx)
if err != nil {
return 0, nil, nil, b.checkError(err)
}
query = query.Must(langQuery)
searchResult, err := b.client.Search().
Index(b.indexerAliasName).
Query(query).
Highlight(
elastic.NewHighlight().
Field("content").
NumOfFragments(0). // return all highting content on fragments
HighlighterType("fvh"),
).
Sort("repo_id", true).
From(start).Size(pageSize).
Do(ctx)
if err != nil {
return 0, nil, nil, b.checkError(err)
}
total, hits, _, err := convertResult(searchResult, kw, pageSize)
return total, hits, extractAggs(countResult), err
}
// Close implements indexer
func (b *ElasticSearchIndexer) Close() {
select {
case <-b.stopTimer:
default:
close(b.stopTimer)
}
}
func (b *ElasticSearchIndexer) checkError(err error) error {
var opErr *net.OpError
if !(elastic.IsConnErr(err) || (errors.As(err, &opErr) && (opErr.Op == "dial" || opErr.Op == "read"))) {
return err
}
b.setAvailability(false)
return err
}
func (b *ElasticSearchIndexer) checkAvailability() {
if b.Ping() {
return
}
// Request cluster state to check if elastic is available again
_, err := b.client.ClusterState().Do(graceful.GetManager().ShutdownContext())
if err != nil {
b.setAvailability(false)
return
}
b.setAvailability(true)
}
func (b *ElasticSearchIndexer) setAvailability(available bool) {
b.lock.Lock()
defer b.lock.Unlock()
if b.available == available {
return
}
b.available = available
}