oss/forgejo
oss/forgejo
1
0
Fork 0
mirror of https://codeberg.org/forgejo/forgejo.git synced 2025-09-15 18:56:59 +00:00
Code Wiki Activity

Optimization of labels handling in issue_search (#4228)

Browse source 
This PR optimizes the SQL query and de-duplicate the labels' ids when generating the query string, on the issue page.

<hr/>

### Background

Some time ago, BingBot and some other crawlers have been putting my instance on its knees with requests containing a lot of label ids, like this one :

```
[07/Aug/2023:11:28:37 +0200] "GET /Dolibarr/sendrecurringinvoicebymail/issues?q=&type=all&sort=&state=closed&labels=1%2c1%2c1%2c1%2c1%2c1%2c1%2c1%2c1%2c1%2c1%2c1%2c1%2c1%2c1%2c1%2c1%2c1%2c1%2c1%2c2%2c10%2c2%2c1%2c1%2c10%2c10%2c7%2c6%2c10%2c10%2c3%2c2%2c1%2c5%2c10%2c1%2c6%2c2%2c7%2c3%2c7%2c6%2c10%2c1%2c10%2c1%2c1%2c7%2c7%2c1%2c1%2c1%2c1%2c10%2c10%2c1%2c2%2c1%2c1%2c1%2c1%2c1%2c1%2c1%2c1%2c1%2c1%2c2%2c1%2c12%2c6%2c6%2c10&milestone=0&project=-1&poster=0 HTTP/1.1" 499 0 "-" "Mozilla/5.0 AppleWebKit/537.36 (KHTML, like Gecko; compatible; bingbot/2.0; +http://www.bing.com/bingbot.htm) Chrome/103.0.5060.134 Safari/537.36"
```

Since each of the label ids implies a join, it grows exponentially expensive for the database engine (at least on PostgreSQL but SQLite suffers a little too).

Thus, this PR proposes two enhancements:

* rewrite the database query to use only one squashed condition,
* deduplicate the label ids when generating the URL.

### Performance comparison

Here are some timings on Postgresql-backed, Forgejo 7.0.4 instances :
```sh
$ time curl -s -o /dev/null "http://localhost:3000/toto/tata/issues?q=&type=all&sort=&labels=19%2c25%2c19%2c25%2c19%2c25%2c19%2c25%2c19%2c25%2c19%2c25%2c19%2c25%2c19%2c25%2c19%2c25%2c19%2c25%2c19%2c25%2c19%2c25%2c19%2c25%2c19%2c25%2c19%2c25%2c19%2c25%2c19%2c25%2c19%2c25%2c19%2c25%2c19%2c25&state=open&milestone=0&project=0&assignee=0&poster=0"

real    0m10,491s
user    0m0,017s
sys     0m0,008s
```
...and with the patch:
```sh
$ time curl -s -o /dev/null "http://localhost:3000/toto/tata/issues?q=&type=all&sort=&labels=19%2c25%2c19%2c25%2c19%2c25%2c19%2c25%2c19%2c25%2c19%2c25%2c19%2c25%2c19%2c25%2c19%2c25%2c19%2c25%2c19%2c25%2c19%2c25%2c19%2c25%2c19%2c25%2c19%2c25%2c19%2c25%2c19%2c25%2c19%2c25%2c19%2c25%2c19%2c25&state=open&milestone=0&project=0&assignee=0&poster=0"

real    0m0,094s
user    0m0,012s
sys     0m0,013s
```

### Annex

This issue was originally proposed to [Gitea](https://github.com/go-gitea/gitea/pull/26460) but didn't get much attention, and I switched to Forgejo in the meantime :)

Reviewed-on: https://codeberg.org/forgejo/forgejo/pulls/4228
Reviewed-by: Earl Warren <earl-warren@noreply.codeberg.org>
Co-authored-by: Chl <chl@xlii.si>
Co-committed-by: Chl <chl@xlii.si>
This commit is contained in:
Chl 2024-06-28 05:11:57 +00:00 • committed by Earl Warren
parent 2121a29f89
commit 544cbc6f01
4 changed files with 80 additions and 16 deletions
Download patch file Download diff file Expand all files Collapse all files

26
models/issues/issue_search.go
View file

@ -6,6 +6,7 @@ package issues
import (
"context"
"fmt"
"strconv"
"strings"
"code.gitea.io/gitea/models/db"
@ -13,6 +14,7 @@ import (
repo_model "code.gitea.io/gitea/models/repo"
"code.gitea.io/gitea/models/unit"
user_model "code.gitea.io/gitea/models/user"
"code.gitea.io/gitea/modules/container"
"code.gitea.io/gitea/modules/optional"
"xorm.io/builder"
@ -116,14 +118,30 @@ func applyLabelsCondition(sess *xorm.Session, opts *IssuesOptions) {
if opts.LabelIDs[0] == 0 {
sess.Where("issue.id NOT IN (SELECT issue_id FROM issue_label)")
} else {
for i, labelID := range opts.LabelIDs {
// deduplicate the label IDs for inclusion and exclusion
includedLabelIDs := make(container.Set[int64])
excludedLabelIDs := make(container.Set[int64])
for _, labelID := range opts.LabelIDs {
if labelID > 0 {
sess.Join("INNER", fmt.Sprintf("issue_label il%d", i),
fmt.Sprintf("issue.id = il%[1]d.issue_id AND il%[1]d.label_id = %[2]d", i, labelID))
includedLabelIDs.Add(labelID)
} else if labelID < 0 { // 0 is not supported here, so just ignore it
sess.Where("issue.id not in (select issue_id from issue_label where label_id = ?)", -labelID)
excludedLabelIDs.Add(-labelID)
}
}
// ... and use them in a subquery of the form :
// where (select count(*) from issue_label where issue_id=issue.id and label_id in (2, 4, 6)) = 3
// This equality is guaranteed thanks to unique index (issue_id,label_id) on table issue_label.
if len(includedLabelIDs) > 0 {
subQuery := builder.Select("count(*)").From("issue_label").Where(builder.Expr("issue_id = issue.id")).
And(builder.In("label_id", includedLabelIDs.Values()))
sess.Where(builder.Eq{strconv.Itoa(len(includedLabelIDs)): subQuery})
}
// or (select count(*)...) = 0 for excluded labels
if len(excludedLabelIDs) > 0 {
subQuery := builder.Select("count(*)").From("issue_label").Where(builder.Expr("issue_id = issue.id")).
And(builder.In("label_id", excludedLabelIDs.Values()))
sess.Where(builder.Eq{"0": subQuery})
}
}
}