> Well today I run it with dmalloc. I'll need some time to handle the > 28 GB of log-file, especially as it apparently contains at least one > large chunk of 0-bytes.
OK, so that log file looks inconspicuous. The reallocations are a few various followed by one series of one permanently growing big one (up to 1792464 bytes; maybe reading in the DB into the internal representation?), a growing small one (growing in 4 bytes steps up to 112 bytes), a huge gap without any reallocations, lots (~55% of all) of reallocations from 257 to something smaller and 4 ones from 99 bytes to 190 bytes). Unfortunately that wasn't the end of the log file, there were still ~2900 other memory actions after it.ยน I'll try something different when the failure occurs again - today it didn't (maybe my timing was wrong), so it probably will be a few days. Best regards, Thomas ยน My (maybe completely wrong ;-) theory would be that dmalloc only logs after the call returned and it failed in the call itself. (I logged using the tags log-trans and log-stats aka 0x9.) -- ๐ฃ๐ฑ๐ธ๐ถ๐ช๐ผ ๐๐ธ๐ป๐ท๐ฎ๐ป
