On 12/10/2015 02:23 PM, Kinkie wrote:
> On Thu, Dec 10, 2015 at 10:07 PM, Alex Rousskov wrote:
>> On 12/10/2015 11:25 AM, Francesco Chemolli wrote:
>>> And it is actually pretty detectable: let’s choose an arbitrary size
>>> of the shared segment and output a warning message to cache.log. This
>>> won’t help with the speed, but it might help with the confusion..


>> 2015/12/09 11:24:51| Total mlock() delay exceeded 5.3 seconds. See
>> shared_memory_locking in squid.conf.documented.


> I was thinking more of something _before_ the delay:
> e.g. if shared memory is > 100mb, then
> "locking XXX mbytes of shared memory. On some systems this may require
> a long time. Squid is not dead, it's just thinking"


I think the post-event warning is better:

* It is difficult to guess which segment sizes will cause significant
delays in a given environment.

* Squid allocates many segments and a few large segments. Warning the
admin before each large allocation would create a lot of noise. Not
warning the admin before some large allocations would kind of defeat the
idea of the preemptive warning itself.

* There is a possibility that the significant (from admin point of view)
startup delay is caused by the _cumulative_ effect of locking various
shared memory segments and not by any single segment.

* Warning folks about something that has not happened yet and may never
happen increases log noise. In this particular case, we _can_ warn when
we know that startup was significantly delayed because of locking,
reducing that noise.

* There is a good chance that somebody concerned about startup delays
will notice the first line printed _after_ the delay.


Are you sure that preemptive per-segment warnings are better than an
acknowledgement of the problem after it happens? Do you think the latter
is likely to be missed by admins investigating startup delays?


> and then when it's done "shared memory locking of X Mb done in Y
> seconds. This may be normal depending on the size of the shared memory
> area".

I would rather put explanations in squid.conf.documented so that we do
not have to risk misleading folks when discussing complex issues using a
one-line statement. Besides, we need to point folks to options
controlling this behavior (in case they do not think that slow startup
is "normal" for them).


Thank you,

Alex.

_______________________________________________
squid-dev mailing list
[email protected]
http://lists.squid-cache.org/listinfo/squid-dev

Reply via email to