You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@openwhisk.apache.org by OpenWhisk Team Slack <ra...@apache.org> on 2022/09/21 15:31:21 UTC

[slack-digest] [2022-09-20] #general

2022-09-20 23:34:56 UTC - Brendan Doyle: just fyi this is a pretty critical bug fix for the scheduler if you're operating it -&gt; <https://github.com/apache/openwhisk/pull/5326>
https://openwhisk-team.slack.com/archives/C3TPCAQG1/p1663716896475219
----
2022-09-20 23:56:49 UTC - Brendan Doyle: Wanted to open a discussion as well on the invoker function cache, already discussed with Dom the general issue but the function is evicted if an activation isn't received by the invoker after 5 minutes. The default idle timeout is longer than that so it's fairly common to hit cases where you have a paused warm container but then have to re-download the function when unpausing. The issue is exacerbated on the new scheduler as function containers are distributed randomly across the invoker fleet rather than the original algorithm that targeted a home invoker so it's less likely to have an invoker go without an activation to refresh the cache.

My original thought was to just change the function cache eviction time to be configurable by collection type. But upon thinking further, maybe it would be safer to have the container proxy periodically ping the cache and if it were to get in a state somehow that the function is evicted, it would re-download the function off the critical path of an activation execution. Thoughts?
https://openwhisk-team.slack.com/archives/C3TPCAQG1/p1663718209874009?thread_ts=1663718209.874009&cid=C3TPCAQG1
----