Retry different disk when bootstrap fails in Full auto mode #2659
Conversation
ambry-store/src/main/java/com/github/ambry/store/DiskManager.java
Outdated
Show resolved
Hide resolved
Arun-LinkedIn
changed the title
Codecov ReportAttention:
Additional details and impacted files@@ Coverage Diff @@
## master #2659 +/- ##
=============================================
- Coverage 69.18% 33.36% -35.83%
+ Complexity 11010 5142 -5868
=============================================
Files 806 809 +3
Lines 65526 65560 +34
Branches 8006 8001 -5
=============================================
- Hits 45337 21875 -23462
- Misses 17620 41991 +24371
+ Partials 2569 1694 -875 ☔ View full report in Codecov by Sentry. |
| throw new StateTransitionException("Failed to add store " + partitionName + " into storage manager", | ||
| ReplicaOperationFailure); | ||
| } else { | ||
| // TODO: Delete any files added in store and reserve directory |
There was a problem hiding this comment.
we should do this, otherwise, it would just take some disk space and won't let it go before we restart. I suppose we can just add a new method in disk manager to clean up the "unexpected dirs". I have a PR to do that, let's merge these two PRs and then use the method in my PR to remove the partition directories later.
There was a problem hiding this comment.
Sure. I will merge this PR and put up a new PR
| @@ -16,6 +16,7 @@ | |||
|
|
|||
| import com.github.ambry.account.AccountService; | |||
| import com.github.ambry.clustermap.DiskId; | |||
| import com.github.ambry.clustermap.HardwareState; | |||
There was a problem hiding this comment.
this import is not used anywhere.
| @@ -113,6 +113,7 @@ public class HelixClusterManager implements ClusterMap { | |||
| private final PartitionSelectionHelper partitionSelectionHelper; | |||
| private final Map<String, Map<String, String>> partitionOverrideInfoMap = new HashMap<>(); | |||
| private final Map<String, ReplicaId> bootstrapReplicas = new ConcurrentHashMap<>(); | |||
| private final Map<String, Set<DiskId>> disksAttemptedForBootstrap = new ConcurrentHashMap<>(); | |||
There was a problem hiding this comment.
one slightly different approach here is avoid using a disk whenver it fails a bootstrap replica. But this also would do the work
Retry different disk when bootstrap fails