[Varnish] #1119: Varnish 3.0.2 freezed, Pushing vcls failed:#012CLI communication error (hdr)

[Varnish]

#1119: Varnish 3.0.2 freezed, Pushing vcls failed:#012CLI communication error
(hdr)
-------------------------------------------------+-------------------------
 Reporter:  campisano                            |       Owner:
     Type:  defect                               |      Status:  closed
 Priority:  normal                               |   Milestone:
Component:  build                                |     Version:  3.0.2
 Severity:  blocker                              |  Resolution:  worksforme
 Keywords:  child died pushing vcls failed       |
  #012CLI communication error (hdr)              |
-------------------------------------------------+-------------------------

Comment (by lampe):

 Replying to [comment:11 campisano]:

 > is the root the master process that handle the 'varnish childs workers'
 ?

 Not quite. As I understand it (haven't actually checked with the code
 yet), the child blocks on a write to the shared memory file when cached
 disk writes exceed a certain amount and the disk is busy. With large RAM,
 the disk sync can take several seconds. If the child blocks long enough,
 the VCL upload from the master process times out and the child is
 terminated but not restarted.

 > Reducing the shm log size on disk can resolve the problem ??

 I don't think so. Putting the shm log file on tmpfs resolved it for me. No
 physical disk, thus no waiting on the I/O scheduler. The Varnish Book
 explicitly recommends that the log must not cause physical disk I/O.

 I'll also experiment with lowering /proc/sys/vm/dirty_background_bytes to
 <1GB and increasing dirty_ratio to 50%. See
 http://www.kernel.org/doc/Documentation/sysctl/vm.txt for a description of
 these linux kernel parameters.

-- 
Ticket URL: <https://www.varnish-cache.org/trac/ticket/1119#comment:12>
Varnish <https://varnish-cache.org/>
The Varnish HTTP Accelerator