Re: store client bug

On Thu, 25 May 2000, Adrian Chadd wrote:

>
>
>
> Why is this code like this:
>
> store_client.c:300
>
> } else if (!sc->flags.disk_io_pending) {
> sc->flags.disk_io_pending = 1;
> storeSwapInStart(sc);
> if (NULL == sc->swapin_sio) {
> storeClientCallback(sc, -1);
> return;
> }
> /*
> * If the open succeeds we either copy from memory, or
> * schedule a disk read in the next block.
> */
>
> Why is the code setting disk_io_pending ? THe problem in the current
> souce tree is that disk_io_pending is set to 1, then a read is scheduled
> at the end of tostClientCopy3() (where this code is also from) and
> triggers an assertion :
>
> /* What the client wants is not in memory. Schedule a disk read */
> assert(STORE_DISK_CLIENT == sc->type);
> assert(!sc->flags.disk_io_pending);
> debug(20, 3) ("storeClientCopy3: reading from STORE\n");
> storeClientFileRead(sc);
>
> I've commented the code out and I can't see any ill effects. Am I
> missing something?

Your fix is correct. It was introduced when I merged
the 2.3 branch to head.

http://www.squid-cache.org/cgi-bin/cvsweb.cgi/squid/src/store_client.c.diff?r1=1.90&r2=1.91

Duane W.
Received on Mon May 29 2000 - 22:56:58 MDT