Botond Dénes
Botond Dénes
> > The checks for when to go into local mode seem a little too loose for me: I was expecting us looking at the replication strategy, although this would...
This page is still valid for partition-range scans and any query having to read from a pre `mc` sstable. So in practice, most users will not hit it.
Also note that 5.0 doesn't have all the reverse query related changes. AFAIK the cache/memtable still didn't support reverse reads.
``` 2022-04-16T05:50:16+00:00 rolling-upgrade-master-centos-db-node-02413cd2-0-3 ! INFO | [shard 1] reader_concurrency_semaphore - Semaphore _read_concurrency_sem with 100/100 count and 2235118/127171297 memory resources: timed out, dumping permit diagnostics: 2022-04-16T05:50:16+00:00 rolling-upgrade-master-centos-db-node-02413cd2-0-3 ! NOTICE | 2022-04-16T05:50:16+00:00...
We won't be able to investigate issues like this until this is fixed.
I see two patterns here: 1) Moderate amounts of reads against `system.auth`: ``` 1 1 16K system_auth.roles/data-query/active/used 53 0 0B system_auth.roles/data-query/waiting ``` These look like single partition reads hitting cache...
I haven't looked at it in detail yet.
@avikivity please review.
I was hesitant to merge because I'm not familiar enough with this code to really understand the change. But in the interest of forward progress I queued it.
> can we piggyback the cache hit/miss status on the replica read / digest reply and use it to update the coordinator's view of the replica's cache? We already do,...