🎉 Snyk checks have passed. No issues have been found so far.

✅ security/snyk check is complete. No issues have been found. (View Details)

✅ license/snyk check is complete. No issues have been found. (View Details)

defrag CI(failed) with commit 0257441 (#14326):
https://github.com/sundb/redis/actions/runs/17460851901/job/49584975468

defrag CI(all tests passed except a known issue) with commit c8fa4bb (#14326)

maybe add another mode in which we run full defrag cycle in each beforeSleep (then we don't even need to fake high fragmentation metric)

@oranagra in the commit, I changed the defrag duty cycle to 30 seconds to avoid failures caused by a large number of slow tests. The effect is roughly the same as manually forcing a full cycle in beforeSleep(). Through this change, I also discovered another issue related to defragmentation.
The main reason is that if we want to add a full fully defrag in beforeSleep, it will bring quite a lot of code changes. I choose a relatively simple way.

Ok. I missed that. That's indeed probably sufficient to scan the whole thing in one cycle.
But I don't currently remember if we added some defragmentation in beforeSleep or is it still only in server serverCron (im AFK).
If it's only in cron, it'll be ineffective (maybe states will get undone before it sees them).
P.s. I suppose we can skip all tests that use large amount of data, and also we can reduce the interval to weekly.

Ok. I missed that. That's indeed probably sufficient to scan the whole thing in one cycle. But I don't currently remember if we added some defragmentation in beforeSleep or is it still only in server serverCron (im AFK). If it's only in cron, it'll be ineffective (maybe states will get undone before it sees them).

We are no longer doing defragmentation in cron now, but in ae timer. So after the commit: 2432cff, As long as the fragment is complete and the cycle does not exceed 30 seconds (if one completed cycle ends, we will immediately start the next one), each ae timer will be triggered.

P.s. I suppose we can skip all tests that use large amount of data, and also we can reduce the interval to weekly.

I tried to comment out the tests that are consuming too much time, but there might be quite a few tests that need to be commented out. Below is the test duration consumption when I used fully defrag. The total test time was one and a half hours.

166 seconds - integration/replication-rdbchannel
  370 seconds - unit/scan
  334 seconds - unit/type/zset
  396 seconds - unit/maxmemory
  419 seconds - unit/other
  386 seconds - unit/type/string
  246 seconds - integration/replication-psync
  461 seconds - unit/lazyfree
  352 seconds - integration/replication-2
  508 seconds - integration/replication
  1359 seconds - unit/sort
  2030 seconds - unit/scripting

ok. so essentially it'll run on each event loop.

regarding duration, that's not too bad..
a better approach could be to check duration of each individual test, then the ones taking very long will stand out (not the units that just happen to have a lot of tests)

from my perspective, you can merge it.

fully CI: https://github.com/sundb/redis/actions/runs/17546998661/job/49830644807

Intentional failures(commit c8876fc) can be detected by this PR.
failed CI: https://github.com/sundb/redis/actions/runs/17576123234/job/49930009235

Run the CI ten times and all passed, i think we can merge it now.