[Bug][Blueprint] Cron-triggered pipelines use a sanitized plan after PATCH /blueprints, gitextractor fails with masked credentials
Mantenedores costumam responder em até 3 dias
Ninguém assumiu esta issue ainda.
Avaliação
- Dificuldade
- 4/5
- Tempo estimado
- 1-2 dias
- Facilidade para iniciantes
- 56/100
Direção de pesquisa
Read backend/server/services/blueprint.go around PatchBlueprint, saveBlueprint, and reloadBlueprint, then inspect SanitizeBlueprint and pluginOptionSanitizers in backend/server/services/pipeline.go. Trace how the cron job and API response share the blueprint plan, then review existing blueprint service tests. Done means cron runs after a PATCH use the stored credentials while the API response still masks them.
Escrita pelo modelo de indexação a partir do texto da issue.
Descrição
Search before asking
- I had searched in the issues and found no similar issues.
What happened
After updating an advanced-mode blueprint through the API (PATCH /blueprints/:id), the next pipeline started by the cron schedule runs with a sanitized (masked) plan. Every gitextractor task fails with:
subtask Clone Git Repo ended unexpectedly
git cmd [git clone https://git:*****@<gitlab-host>/<group>/<repo>.git /tmp/gitextractorXXXX --bare] failed
remote: HTTP Basic: Access denied. ...
The token stored in the database is correct:
- Triggering the same blueprint manually (
POST /blueprints/:id/trigger) clones every repo fine. - After a
lakerestart, cron runs work again until the next PATCH.
In our case it was 1330/1330 gitextractor tasks failing in a cron run, while the same repos cloned fine in a manual run of the same plan a few hours earlier.
Root cause (v1.0.3-beta15, still present on main):
PatchBlueprint→saveBlueprint→reloadBlueprint(blueprint)registers the cron job with the same pointer:
cronManager.AddJob(blueprint.CronConfig, &BlueprintJob{blueprint})(backend/server/services/blueprint.go).- Right after saving,
PatchBlueprintcallsSanitizeBlueprint(blueprint)on that same object to build the API response. - The
gitextractorsanitizer (pluginOptionSanitizersinbackend/server/services/pipeline.go) replaces the password inoptions["url"]with*in place. - When the cron fires,
BlueprintJob.Run()callscreatePipelineByBlueprint(bj.Blueprint, ...), which for advanced mode usesblueprint.Planas is. The new pipeline gets the masked URL.
TriggerBlueprint and ReloadBlueprints (at startup) load the blueprint from the DB, which is why manual runs and runs after a restart are fine. Normal-mode blueprints are not affected, because their plan is regenerated from the connections on each run.
What do you expect to happen
Cron-triggered pipelines should use the stored (unsanitized) plan, the same as manual triggers. Sanitizing the API response should not change the object the scheduler holds.
How to reproduce
- Create an advanced-mode blueprint with a
gitextractortask whoseurlcontains credentials (https://git:<token>@host/group/repo.git), withenable: true,isManual: falseand a cron, e.g.*/5 * * * *. - Update it through the API, e.g.
PATCH /blueprints/:idwith{"cronConfig": "*/5 * * * *"}or with a newplan. - Wait for the cron to fire.
- The
gitextractortask fails withHTTP Basic: Access denied, and the pipeline's task options show the masked URL. - Trigger the same blueprint with
POST /blueprints/:id/trigger: it succeeds.
Anything else
Possible fixes, either of:
- In
BlueprintJob.Run(), reload the blueprint from the DB by id before creating the pipeline. - In
PatchBlueprint(and other handlers that sanitize aftersaveBlueprint), sanitize a deep copy rather than the object passed toreloadBlueprint.SanitizeBlueprintwrites intoblueprint.Plan[i][j], and the option maps are shared.
Workaround: set the blueprint to manual and trigger it from an external scheduler with POST /blueprints/:id/trigger, or restart lake after every blueprint update.
Version
v1.0.3-beta15 (helm chart 1.0.3-beta15). Code verified unchanged on main.
Are you willing to submit PR?
- Yes I am willing to submit a PR!
Code of Conduct
- I agree to follow this project's Code of Conduct
- Linguagem predominante
- Go
- Estrelas
- 3.2k
- Forks
- 824
- Merge médio
- 2d 22h
- PRs com merge (30d)
- 46
Preparar o ambiente
Inicia o contêiner de desenvolvimento do projeto no navegador, com a sua própria conta do GitHub.
- Sem Dockerfile nem arquivo Docker Compose
- Tem um modelo de pull request
- Sem guia de contribuição
Primeiros passos
- Leia a issue inteira e depois o guia de contribuição do projeto.
- Comente na issue dizendo que vai assumir — evita que duas pessoas façam o mesmo trabalho.
- Faça um fork do repositório e trabalhe em uma branch.
- Abra um pull request que referencie o número da issue.
Mais de apache/devlake
-
[Bug][SonarQube] Migration 20230927145127 is not re-runnable: fails with "constraint _tool_sonarqube_file_metrics_pkey does not exist" after an interrupted runTalvez já em andamento @chenwei791129 assumiu há 1 dia. Aberta
Dificuldade 3/5 1-2 dias Facilidade para iniciantes 35/100
Mantenedores costumam responder em até 3 dias
-
type/bug
Dificuldade 3/5 1-2 dias Facilidade para iniciantes 64/100
Mantenedores costumam responder em até 3 dias
-
[Bug][Blueprint] Scheduled run fails silently when plan generation errorsTalvez já em andamento @jcamargos assumiu há 2 dias. Aberta
Dificuldade 4/5 3-5 dias Facilidade para iniciantes 25/100
Mantenedores costumam responder em até 3 dias
-
Dificuldade 3/5 Meio dia Facilidade para iniciantes 62/100
Mantenedores costumam responder em até 3 dias
-
Dificuldade 3/5 1-2 dias Facilidade para iniciantes 42/100
Mantenedores costumam responder em até 3 dias
Todas as issues de apache/devlake
Issues semelhantes
-
Dificuldade 2/5 1-3 horas Facilidade para iniciantes 76/100
Mantenedores costumam responder em até 1 dia
-
Dificuldade 2/5 1-3 horas Facilidade para iniciantes 82/100
-
Dificuldade 2/5 1-3 horas Facilidade para iniciantes 78/100
-
bug
Dificuldade 2/5 1-3 horas Facilidade para iniciantes 78/100
Mantenedores costumam responder em até 2 dias
-
bug
Dificuldade 2/5 1-3 horas Facilidade para iniciantes 83/100
devlikeapro/waha#2322 ·
Mantenedores costumam responder em até 2 dias