100% CPU use in mhttpd

dd1 reporter

specific problem we saw in CERN ALPHA-2 was caused by a history corruption problem in the Summer 2021 where duplicate history files were somehow created (a problem with system time? I forget). this confused this history reader, duplicated schema was correctly removed, but it forgot that it already read the corresponding files, so it was re-reading them again and again. fix is to keep track of already-read files explicitly instead of through the schema list. with this fix, we no longer see 100% CPU usage in mhttpd.

commit 2740aa60c0ef292cea91461c5e60fe8b6c75fffd

K.O.

2022-05-17T16:13:37+00:00

Comments (3)

Stefan Ritt
After we discussed the problem last time, I tried many things. The one which really worked was to buy a new SSD disk and run the history there.
- 2022-05-04T19:31:00+00:00
dd1 reporter
specific problem we saw in CERN ALPHA-2 was caused by a history corruption problem in the Summer 2021 where duplicate history files were somehow created (a problem with system time? I forget). this confused this history reader, duplicated schema was correctly removed, but it forgot that it already read the corresponding files, so it was re-reading them again and again. fix is to keep track of already-read files explicitly instead of through the schema list. with this fix, we no longer see 100% CPU usage in mhttpd.

commit 2740aa60c0ef292cea91461c5e60fe8b6c75fffd

K.O.
- 2022-05-17T16:13:37+00:00
dd1 reporter
- changed status to resolved
fixed. K.O.
- 2022-05-17T16:13:47+00:00
Log in to comment