Solved!
Thanks for the reply, jmcqueen. That was one of the first things I checked, and didn't find anything usual. As far as Apache was concerned, nothing out of the ordinary was happening. It was still able to serve the files, and all gave 200s for the downloads. It all just happened very, very slowly. Even putting Apache's logging in debug mode did not show anything strange.
Also, I forgot to mention another detail regarding the strange behavior. For the big files, it would start out at full speed, then at some point, the download would suddenly slow down to a crawl. This point in the file download was consistent, even with retries.
I talked to a senior colleague of mine and he said that he saw the same behavior before, but with the FTP service of a box running another distro. He said after days of searching for the source of the issues, he found that the XFS volumes' journaling had gone bad. My server is using ext3, but it was my only lead.
I forced an fsck, and so far the issue has not resurfaced! I wasn't able to see the results of the fsck, since I left it running and when I came back all I could see was the login GUI (and apparently fsck doesn't write logs about what it found, presumably because the volumes are unmounted during the process). So far though, everything is looking good.
Hope somebody finds this information useful!
|