On 12/04/2015 08:37 AM, Hussam Al-Tayeb wrote: > Since this is a database, it is possible for part of the database to > get corrupted through a crash or incorrect poweroff? It depends on your definition of "corruption". Yes, it is possible that some database updates will be incomplete because of a poweroff. Bugs notwithstanding, after a restart, * Squid will not notice that an entry that was supposed to be deleted was not. Squid will continue to serve such an entry from the cache. * Assuming atomic single-write disk I/Os, Squid should notice an entry that was only partially saved and not serve it from the cache. Its slots will be considered free space. * In the event a single-write disk I/O was only partially completed, Squid may or may not notice a partial save, depending on what was actually written to disk. There is currently no Squid code that detects non-atomic single-write disk I/Os. AFAICT, this might corrupt up to two cache entries per cache_dir in such a way that Squid will not notice the corruption unless there are some OS-level protections against that. Squid uses regular file system calls for writing entries... * There should be no effect on entries already fully stored at the time of the power outage. > I had an incorrect poweroff yesterday but cache.log did not list > anything weird. > Nevertheless, what would be the best way to check if there was some > damage to the database (unusable slots/cells/whatever)? IIRC, for Rock, all validation is currently done automagically upon startup. HTH, Alex. _______________________________________________ squid-users mailing list squid-users@xxxxxxxxxxxxxxxxxxxxx http://lists.squid-cache.org/listinfo/squid-users
