On Thu, Mar 27, 2008, Neil Harkins wrote: > As for Squid handling a JBOD single disk failure, > not stacking up more reads on an (assumed) failed > disk would be great, but the process still needs to > be killed to get rid of those that blocked before it > noticed and to replace the disk, right? There's a few things to do here: * fail existing requests and new requests! * mark the cachedir as down; stop any hits from that cachedir * (delete the StoreEntry records for that particular FS) * modify the rebuild logic to handle rebuilding a single FS when it comes online! * (And if possible - make the rebuild logic for ufs/aufs/diskd use the async operators instead of the sync operators) Ideally the index should be pushed into per-storedir (so the whole index can be removed all at once!) or place the StoreEntry entries in a linked list per storedir - both of which are to avoid walking the whole tree just to delete the entries for one storedir. Adrian -- - Xenion - http://www.xenion.com.au/ - VPS Hosting - Commercial Squid Support - - $25/pm entry-level VPSes w/ capped bandwidth charges available in WA -
