-
Type: Bug
-
Resolution: Gone away
-
Priority: Major - P3
-
None
-
Affects Version/s: 4.4.18
-
Component/s: None
-
None
-
Replication
-
ALL
I just tried upgrading my RS running 4.4.18 from fCV 4.2 to fCV 4.4 (yes it has been running as 4.2 for a white ^^). Unfortunately bad surprise, I immediately saw the disk IO load (write) on both secondary jump to the roof (about 8 times from ~12% to ~90%). The amount of data written only slightly increased though, so it's mostly the number of writes, which are now smaller I suppose. There's also a ×2 increase in the CPU usage and a small network usage increase, probably related. As soon as I downgraded back to fCV 4.2, the IO, CPU and network load went down back to previous levels (times are UTC+1 here):
CPU:
I checked of course the features which requires fCV 4.4 but there's only two listed here and I don't see any likely culprit: https://www.mongodb.com/docs/manual/release-notes/4.4-compatibility/#std-label-4.4-compatibility-enabled
So I suppose there are other changes caused by this flag, are you aware of any change that could cause this huge regression? Maybe there's a configuration default which changes that I could tune?
I have the diagnostic.data and logs for these servers if you let me know where I can upload them safely.
I wanted to do this in order to prepare the upgrade to 5.0 and 6.0, but I can't really continue with the upgrade or try newer versions if 4.4 already puts me in such a difficult position. Also sorry if there's already a ticket about this, I did try to search but couldn't find any.
Thanks!
- is related to
-
SERVER-54939 Investigate secondary batching behavior in v4.4
- Closed