- 12 May, 2014 5 commits
-
-
Poul-Henning Kamp authored
(Remember to change the "PURGE" to "GET" if you do this.) Fixes #1493
-
Martin Blix Grydeland authored
Fixes: #1499
-
Poul-Henning Kamp authored
-
Poul-Henning Kamp authored
reserved C keywords. Fixes #1498
-
Poul-Henning Kamp authored
object length while processing the Range header to ensure consistent results when streaming from the backend.
-
- 08 May, 2014 1 commit
-
-
Federico G. Schwindt authored
Submitted by: github::matsuu While here add missing semicolon.
-
- 06 May, 2014 9 commits
-
-
Per Buer authored
-
Per Buer authored
-
Poul-Henning Kamp authored
-
Lasse Karstensen authored
Add support for using this if it is defined by the administrator in the /etc/sysconfig/varnish file. From way back we've had a commented-out DAEMON_COREFILE_LIMIT in our /etc/sysconfig/varnish file, which wasn't referenced anywhere. Suggested by: gaoyongwei/itxx00 on github.
-
Lasse Karstensen authored
-
Poul-Henning Kamp authored
Suggested by: Steven Engelhardt via github
-
Martin Blix Grydeland authored
-
Martin Blix Grydeland authored
This to give predictable backend reuse behavior for varnishtest, making the test cases non-racy. 2nd version of this patch. This version will only deal with recyclable backend connections before BOS_FINISHED, leaving the slower close path to the clean up code that is run after the BOS_FINISHED state. Fixes: #1491
-
Poul-Henning Kamp authored
instances whenever we give the server instance more work, doing so runs into a dark and murky corner somewhere in Stevens and Solaris sometimes doesn't like that. The trouble is that the initial bind(2) asks the kernel to assign a port, but the subsequent insist on reusing that port, and that, in some cases, could mean toruble which cannot be allowed. Add a "-break" instruction to server so that test case c00035 can still work. Spotted by: Nils Goroll / UPLEX
-
- 05 May, 2014 19 commits
-
-
Nils Goroll authored
-
Nils Goroll authored
-
Nils Goroll authored
-
Nils Goroll authored
-
Lasse Karstensen authored
-
Nils Goroll authored
-
Joshua Bussdieker authored
-
Lasse Karstensen authored
Update changelog
-
Poul-Henning Kamp authored
Submitted by: github::brianjarita
-
Lasse Karstensen authored
As discussed on bugwash today. Lowering the bar may help us improve the documentation.
-
Dag Haavi Finstad authored
-
Nils Goroll authored
-
Martin Blix Grydeland authored
The pool herder didn't signal the thread when decimating the flock, causing the thread to be leaked. Spotted by: xcir Fixes: #1490 Also close a couple of other races: When pthread_cond_timedwait returns ETIMEDOUT, the predicate (wrk->task.func == NULL) could still have changed while reacquiring the mutex. If so, the signal would've been lost and the thread leaked. Changed the idle wait loop in Pool_Work_Thread() to always check the predicate before resuming the cond_wait. The update of the predicate (wrk->task.func) from e.g. Pool_Task() is done without holding the mutex. This opens a race on the predicate, and the possibility of the worker going back on cond_wait loosing the signal. Changed to update the predicate while holding the mutex.
-
Poul-Henning Kamp authored
Spotted by: Coverity
-
Poul-Henning Kamp authored
-
Poul-Henning Kamp authored
Spotted by: Coverity
-
Poul-Henning Kamp authored
-
Per Buer authored
-
Poul-Henning Kamp authored
In 4.x we changed obj.hits to be read-only, and to count all variants under the same hash-key (= same objhdr). Since we cannot trivially tell if a ban hits all Vary: values or just some of them, there is no sane heuristic to fiddle obj.hits on bans that would Do The Right Thing. Fixes #1492
-
- 02 May, 2014 1 commit
-
-
Lasse Karstensen authored
Also remove the libvarnishapi "workaround" we put in place earlier. Found by: Ingvar Hagelund Fixes: #1484
-
- 30 Apr, 2014 1 commit
-
-
Martin Blix Grydeland authored
Fixes: #1489 Spotted by: xcir
-
- 28 Apr, 2014 2 commits
-
-
Federico G. Schwindt authored
Update board members, URLs, spelling and fmt(1).
-
Martin Blix Grydeland authored
The disembarking thread would clear the req->wrk pointer in CNT_Request. The req could have already been rescheduled on another worker at this point, causing the req->wrk to be cleared while processing. Change back to not clearing the pointer for REQ_FSM_DISEMBARK (it's already been done under mutex in HSH_Lookup anyways). Fixes: #1488
-
- 27 Apr, 2014 1 commit
-
-
Federico G. Schwindt authored
-
- 25 Apr, 2014 1 commit
-
-
Martin Blix Grydeland authored
backend IMS. When revalidating using backend IMS, copy the status code, status code, status string and response message from the original object into the new revalidated object. This makes sure that none of the 304 message fields gets applied to the new revalidated object. Fixes: #1485
-