Christian Schneider created SLING-13356:
-------------------------------------------
Summary: ResourceDistributionPackageCleanup deletes all disposable
packages in a single unbounded JCR commit
Key: SLING-13356
URL: https://issues.apache.org/jira/browse/SLING-13356
Project: Sling
Issue Type: Bug
Affects Versions: Content Distribution Core 0.7.4
Reporter: Christian Schneider
Assignee: Christian Schneider
Fix For: Content Distribution Core 0.8.0
{{ResourceDistributionPackageCleanup.run()}} iterates every disposable package
and calls {{delete(false)}} (no commit) on each one inside the loop — the only
{{commit()}} happens once, after the whole loop finishes. On a
resource-persisted distribution agent with a large package backlog, this
becomes a single unbounded transaction.
We saw a real case where a distribution agent accumulated a backlog of roughly
1.5-1.75 million disposable packages. The resulting single commit (~7 million
node deletions) never completed and put the underlying JCR repository under
sustained load until the backlog was cleared out-of-band.
Proposed fix: commit in bounded batches inside the loop instead of once at the
end — track packages deleted since the last commit, and call {{commit()}} once
a configurable batch size is reached (falling back to today's
single-commit-at-the-end behavior when unset or ≤ 0). Will attach a patch.
Related: SLING-10743 — same class, same package-path pattern in its own
example, but a different root cause (concurrent-instance arbitration, already
fixed).
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
