Don't ignore the NotFoundErr when handling the status hook of tasks to avoid the status out of sync
Fixes#14016
Signed-off-by: Wenkai Yin <yinw@vmware.com>
- update StopJob() of basic worker
- update UnSchedule() of basic scheduler
- update the policy store to get more data
fix#13599 , fix#13597
Signed-off-by: Steven Zou <szou@vmware.com>
Remove the artifact trash record after manifest delete success.
Don't use the flush method is because that when all of records are removed, only GC job knows these informations, and they are in the memory. Once the jobservice is crashed and restarted at GC job execution phase, the trash records are lost, then these manifest are become orphan manifests, cannot be removed any more.
Signed-off-by: wang yan <wangyan@vmware.com>
Add the read only job as a back up plan, user still can use it but just with API, and specify the parameter read_only:true
Signed-off-by: wang yan <wangyan@vmware.com>
1, update typo in the update blob status sql, the typo will not impact the sql result.
2, correct blob status in the middleware & GC job log.
Signed-off-by: wang yan <wangyan@vmware.com>
To handle the orphan blobs that created in the quota exceeding case deletion, remove the limition of artifact remove.
Signed-off-by: wang yan <wangyan@vmware.com>
* add debugging env for GC time window
For debugging, the tester/users wants to run GC to delete the removed artifact immediately instead of waitting for two hours, add the env(GC_BLOB_TIME_WINDOW) to meet this.
Signed-off-by: wang yan <wangyan@vmware.com>
two phases:
1, mark, select the gc candidates bases on the DB and mark them as status delete.
2, sweep, select the candidate and mark it as status deleting and remove it from backend and database.
Signed-off-by: wang yan <wangyan@vmware.com>
- define policy enforcer interface
- implement the default enforcer
- registrer P2P preheat job to JS
- add the missing mock manager&controller in the src/testing pkg
- Add UT cases for enforcer
- fix#12285
- left one TODO: query provider instance by instance Manager
Signed-off-by: Steven Zou <szou@vmware.com>
1, add two more attributes, update_time and status
2, add delete and fresh update time method in blob mgr & ctr.
Signed-off-by: wang yan <wangyan@vmware.com>
- returnning nil instead of error when trying to stop a job that has been in the final status(Error/Success/Stopped)
- do enhancements to the periodic job unschedule func
- fix a UT nil ptr issue
Signed-off-by: Steven Zou <szou@vmware.com>
- never expire the jobs that are not entering the final status (Error,Success or Stopped)
- set different expireation time to the jobs with different status
- never store the `check_in` data in the redis db to save space
Signed-off-by: Steven Zou <szou@vmware.com>
Fixes#9704
As we do want to unify error handling, so just decreprates pkg errors, use lib/errors instead for Harbor internal used errors model.
1, The lib/errors can cover all of funcs of pkg/errors, and also it has code attribute to define the http return value.
2, lib/errors can give a OCI standard error format, like {"errors":[{"code":"UNAUTHORIZED","message":"unauthorized"}]}
If you'd like to use pkg/errors, use lib/errors instead. If it cannot meet your request, enhance it.
Signed-off-by: wang yan <wangyan@vmware.com>
- priority option is supported when doing job registration
- the priority is defined by a unique priority sampler
- the default priority is 1000 (max is 10000)
Signed-off-by: Steven Zou <szou@vmware.com>
- update Job interface to introdcue MaxCurrency method for declaring the max currency of the specified job
- change the downstream jobs to implement the new interface method
- GC and sample jobs are set to 1
- other jobs are set to 0 that means unlimited
- add max currency optiot when doing job registration
- resolve issue #11586
- probably resolve issue #11281
- resolve issue #11570
Signed-off-by: Steven Zou <szou@vmware.com>
fixes#11533
GC jobs will use the filter results to call registry API to delete manifest.
In the current imple, the filter function in some case does not return the deleted artifact as it's using digest as the filter condition.
Like: If one artifact is deleted, but there is another project/repo has a image with same digest with the deleted one, filter func will
not mark the deleted artifact as candidate. It results in, GC job does not call API to remove the manifest.
To fix it, update the filter to use both digest and repository name to filter candidate.
Signed-off-by: wang yan <wangyan@vmware.com>
- schedule the periodical jobs following the UTC timezone
- e.g: 5 10 10 * * * means run jobs at UTC time 10:10:05 everyday
- fix issue #11466
Signed-off-by: Steven Zou <szou@vmware.com>
Fixes#11190, delete all of non-referenced blobs of each project in GC job, thun the quota
can be released.
Signed-off-by: wang yan <wangyan@vmware.com>
