tree 8e6b92d3f1b16893d8d31c7c397eb84a1d40184d
parent fe30fb8b5f37732e3690a3c521d51eb851e73b5c
author Marcin Czech <maczech@gmail.com> 1603836706 +0100
committer Luca Milanesio <luca.milanesio@gmail.com> 1604612063 +0000

Lock indexing events on forwarder site

Before this change, locking was happening on the receiver site. Long
running indexing tasks can cause socket timeout on forwarder site.
Forwarder uses retry mechanism to deliver the event again. Every retry
operation starts a new thread on receiver site. Because locking
happens on receiver each new event for the same id cause more retries
and more threads waiting to acquire the lock. This cause threads
starvation on the receiver site and brings down the Gerrit node.
Adding locking on the forwarder side allows to create a per id lock for
the indexing operations. Even if we have multiple events for a single
id only one is running, others are waiting in the queue or are dropped.

Bug: Issue 13561
Change-Id: I4c603c8f40e94f543e7dcb7b34159d4ca9df497d