Thanks for your efforts in making this change for this SRU cycle.
However, I can confirm that we too are facing challenges with orders of magnitude (>10x) increases in ACCESS calls, with the newly released kernels. Still better/differently than 5.15.0-67, but not suitable for production.
Kernel 5.15.0-69 behaves significantly different than test kernel 5.15.0-67.74+test20230307b2h2cbb6062f8eb. I do not think 5.15.0-67.74+test20230308b0h1a13a615ee3 was ever available from your PPA ?. We ran 5.15.0-67.74+test20230307b2h2cbb6062f8eb for 2 weeks, without issues.
What looks like a similar (apache2) environment to Jan and Ben, we have short(er) lived procs with suexec-like setup relying heavily on caching using nocto and increased AC-timeouts options.
While I understand the reasoning for avoiding stale access caches, and clearing it in relation to group-member changes, I do not think it was ever the intention to break default long-time behaviour.
As of now I cannot confirm issues with increased LOOKUPs in our end as our caches are still warming.
Hi Du ChengEn,
Thanks for your efforts in making this change for this SRU cycle.
However, I can confirm that we too are facing challenges with orders of magnitude (>10x) increases in ACCESS calls, with the newly released kernels. Still better/differently than 5.15.0-67, but not suitable for production.
Kernel 5.15.0-69 behaves significantly different than test kernel 5.15.0- 67.74+test20230 307b2h2cbb6062f 8eb. I do not think 5.15.0- 67.74+test20230 308b0h1a13a615e e3 was ever available from your PPA ?. We ran 5.15.0- 67.74+test20230 307b2h2cbb6062f 8eb for 2 weeks, without issues.
What looks like a similar (apache2) environment to Jan and Ben, we have short(er) lived procs with suexec-like setup relying heavily on caching using nocto and increased AC-timeouts options.
While I understand the reasoning for avoiding stale access caches, and clearing it in relation to group-member changes, I do not think it was ever the intention to break default long-time behaviour.
As of now I cannot confirm issues with increased LOOKUPs in our end as our caches are still warming.