Solve race condition when several calls to apoc.trigger.install are made concurrently. #875
Conversation
| try { | ||
| node = tx.createNode(SystemLabels.ApocTriggerMeta); | ||
| node.setProperty(SystemPropertyKeys.database.name(), databaseName); | ||
| } catch (ConstraintViolationException e) { |
There was a problem hiding this comment.
Is this a new transaction or the outer one? because if it is the outer one won't that cause issues for the user later on in the query?
There was a problem hiding this comment.
Good point, I did not consider this. It looks like the tx is created in withUpdatingTransaction() which is called in the beginning of apoc.trigger.install, apoc.trigger.drop, apoc.trigger.dropAll, apoc.trigger.start, apoc.trigger.stop respectively.
The thread does some work before this method, but it is always commited directly after it. Does that sounds OK?
There was a problem hiding this comment.
Okay, that should be fine then I think
…ade concurrently. Also update some licenses after new Neo4j release.
Lojjs
force-pushed
the
surf-419-apoctriggerinstall-fails-due-to-multiple-apoctriggermeta
branch
from
d35d998 to
94e028d
Compare
| // This can happen if two threads try to create the same node concurrently, | ||
| // after both having passed the null check. In this case we can ignore the failing tx. |
There was a problem hiding this comment.
Perhaps at least one retry? Failing to update lastUpdated can result in that the change to triggers will never propagate (I think).
There was a problem hiding this comment.
We could also make it safer by changing:
if (getLastUpdate() >= lastUpdate) {
updateCache();
}to something like
if (getLastUpdate() >= lastUpdate || lastUpdate < reallyOld) {
updateCache();
}
Lojjs
force-pushed
the
surf-419-apoctriggerinstall-fails-due-to-multiple-apoctriggermeta
branch
from
3ec4097 to
2dad37e
Compare
Lojjs
deleted the
surf-419-apoctriggerinstall-fails-due-to-multiple-apoctriggermeta
branch
Also update some licenses after new Neo4j release.
Should be cherry-picked to 5.26.