gianm commented on a change in pull request #8644: Fix Kinesis resharding issues
URL: https://github.com/apache/incubator-druid/pull/8644#discussion_r332836956
##########
File path:
extensions-core/kinesis-indexing-service/src/main/java/org/apache/druid/indexing/kinesis/supervisor/KinesisSupervisor.java
##########
@@ -212,14 +225,47 @@ protected void
scheduleReporting(ScheduledExecutorService reportingExec)
// not yet implemented, see issue #6739
}
+
+ /**
+ * We try to parse the shard number of the shard ID, using a BigInteger
because the Kinesis shard ID can be
+ * up to 128 characters. The shard number is used preferentially because it
provides a fixed and easily predictable
+ * mapping from shard to task group number.
+ *
+ * If we can't parse the shard number from the ID, then we fall back to
hashing the shard ID string.
Review comment:
I don't think it can happen today, given how Kinesis names shards. It might
happen in the future if they change the format of a shard identifier for some
reason. I don't think that's likely to happen, since it would probably break a
ton of stuff so Amazon would hesitate to do it. So just doing the BigInteger
path only (and throwing an exception if they can't be parsed for some reason)
is a risk but a small one. On the other hand it's zero risk to treat them as
opaque strings.
I read the code in the PR as trying to have it both ways, a 'nice' behavior
(assign shards round-robin based on shard number) but also zero risk of Amazon
doing an unexpected change and breaking us.
Fwiw I think I would go with using the zero risk SHA path only — it would
mean shard assignment that isn't as human-predictable, but it shortens up the
code while still avoiding forward-compatibility risk, which seems worth it to
me.
In the end I am inclined to leave it up to the author @jon-wei.
----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
[email protected]
With regards,
Apache Git Services
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
