[ https://issues.apache.org/jira/browse/CASSANDRA-19529?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17836052#comment-17836052 ]
Nicolas Henneaux edited comment on CASSANDRA-19529 at 4/11/24 8:37 AM: ----------------------------------------------------------------------- I made some tracing and restart of the nodes executing the queries at the same moment after restart. I don't see anything very conclusive anymore. Moreover the graph is not very conclusive anymore despite a higher latency during warm up but this is expected since using LOCAL_QUORUM. I suspect the before was with a long running cluster and after restart it takes some hours to warm-up everything. !screenshot-4.png! On another subject, I realize while playing with this change that LOCAL_QUORUM might be a risky setup in my situation. The setup is spread across 3 physical DC's with two close proximity (<1ms) DC's and one at 10ms. Two Cassandra DC's are setup,one for close proximity physical DC's (C*-DC1) and another for the other DC (C*-DC2). C*-DC1 has 4 nodes with replication of 4 and C*-DC2 with replication of 1. They are working in QUORUM. It allows the loss of two nodes without impact on the availability. With auth using LOCAL_QUORUM, it does not allow such fail-over scenario. was (Author: nicolas.henneaux): I made some tracing and restart of the nodes executing the queries at the same moment after restart. I don't see anything very conclusive anymore. Moreover the graph is not very conclusive anymore despite a higher latency during warm up but this is expected since using LOCAL_QUORUM. I suspect the before was with a long running cluster and after restart it takes some hours to warm-up everything. !screenshot-4.png! > Latency regression on 4.1 comparing to 4.0 > ------------------------------------------ > > Key: CASSANDRA-19529 > URL: https://issues.apache.org/jira/browse/CASSANDRA-19529 > Project: Cassandra > Issue Type: Bug > Components: Feature/Authorization > Reporter: Nicolas Henneaux > Priority: Normal > Fix For: 4.1.x, 5.0.x, 5.x > > Attachments: screenshot-1.png, screenshot-2.png, screenshot-3.png, > screenshot-4.png > > > When upgrading from Cassandra 4.0.10 to 4.1.3, I noticed an increase from > application point of view latency from ~8ms to ~15ms when upgrading to > Cassandra. The latency includes 3 simple queries (INSERT + SELECT (PK+CK) + > UPDATE) plus application overhead. > It has been investigated in CASSANDRA-18766 to realize it is not related. > I tested to downgrade to 4.1alpha1 and the latency regression is still there > with same value. > The version 4.1.4 has the same issue. > In a graph how it looks like > !screenshot-1.png! -- This message was sent by Atlassian Jira (v8.20.10#820010) --------------------------------------------------------------------- To unsubscribe, e-mail: commits-unsubscr...@cassandra.apache.org For additional commands, e-mail: commits-h...@cassandra.apache.org