pycs.net, still broken
Argh - just logged in to the server to find that the pycs.net process has used 1338 minutes (just over 22 hours) of CPU time since Friday, and that it was currently sucking 97% CPU. Ouch. Sorry to all the other users of the JohnCompanies FreeBSD box it's been abusing.
It looks like something is causing MetaKit (the database) to spin while trying to access something. Definitely what happend ages back. At that time, it was the comment database that was all screwed up. Nowadays, PyCS stores a lot more data in there, so it could be anything -- user data, referrer info, comments, trackbacks, access rules, etc.
I'm going to pick up my MetaKit database compaction script from the other day and get it to look a bit deeper into the database, and actually copy each bit of data separately. I suspect that picking up a row and dropping it another file was just copying bytes from DB to DB rather than redoing the structure. Let's see.
Update: Finished the script and ran it on the server. Right now pycs.net is running with the new database, and it looks OK. Let's see how it holds up. Fingers crossed -- hopefully it's not some weird MetaKit/FreeBSD bug...
Update 2: Yup, still going, several hours later. 2 CPU minutes to serve a few thousand pages; seems reasonable.
Update 3: The server was down for a few minutes earlier -- when it came back everything had been restarted, so I assume that it had been rebooted. PyCS is still running fine, though.
... more like this: [Python Community Server]
... topic exchange: [python_community_server]
It looks like something is causing MetaKit (the database) to spin while trying to access something. Definitely what happend ages back. At that time, it was the comment database that was all screwed up. Nowadays, PyCS stores a lot more data in there, so it could be anything -- user data, referrer info, comments, trackbacks, access rules, etc.
I'm going to pick up my MetaKit database compaction script from the other day and get it to look a bit deeper into the database, and actually copy each bit of data separately. I suspect that picking up a row and dropping it another file was just copying bytes from DB to DB rather than redoing the structure. Let's see.
Update: Finished the script and ran it on the server. Right now pycs.net is running with the new database, and it looks OK. Let's see how it holds up. Fingers crossed -- hopefully it's not some weird MetaKit/FreeBSD bug...
Update 2: Yup, still going, several hours later. 2 CPU minutes to serve a few thousand pages; seems reasonable.
Update 3: The server was down for a few minutes earlier -- when it came back everything had been restarted, so I assume that it had been rebooted. PyCS is still running fine, though.