On 27/12/2012 4:52 a.m., carteriii wrote:
I came back to these discussions because I have 3.2.3 and see some excessive
memory usage and found this discussion. I don't know enough to know if my
situation is the same, but it feels like a memory leak and seems relevant to
this discussion. From the cachemgr.cgi "info" page, I found the "space in
area" vs. "total in use" numbers particularly confusing:
Memory usage for squid via mallinfo():
Total space in arena: 1360760 KB
Ordinary blocks: 1360469 KB 25 blks
Small blocks: 0 KB 0 blks
Holding blocks: 25360 KB 6 blks
Free Small blocks: 0 KB
Free Ordinary blocks: 291 KB
Total in use: 291 KB 0%
Total free: 291 KB 0%
Total size: 1386120 KB
Memory accounted for:
Total accounted: 1305182 KB 94%
memPool accounted: 1305182 KB 94%
memPool unaccounted: 80938 KB 6%
memPoolAlloc calls: 7170387
memPoolFree calls: 7170578
If I'm not mistaken, this says squid is consuming over 1GB of memory but
only 291 KB are being used. memPool says that 94% is "accounted", so
wouldn't that imply this is not a memory leak? Or is it a memory leak and
squid knows about it?
There is 1.3 GB allocated to Squid this is agreed by mallinfo and
memPools. ~50MB of that is being tracked by MemPools so is technically
'free' but only for Squid usage.
Not sure what mallinfo is going on about that 291KB for, it is less than
the amount mallinfo is tracking, if this is a 64-bit system I would put
it down to the well-known 32-bit wrap bugs in mallinfo.
It is memory use and Squid knows about it (both agree on the total size,
leak is usually disagreement with mallinfo larger). There is no
identification from the report whether it is leaked internally to Squid
or actually used by some transaction or data cache.
Am I reading this correctly? This is from a system
with light usage that has only been running about a day. Prior to getting
those statistics, I did set these two options in my configuration file to
try to keep the usage down, but it didn't seem to help:
cache_mem 128 MB
This is total size of the RAM object storage, equivalent to a cachr_dir
in memory. Squid uses a lot of additional memory, so this is more like a
minimum than a maximum of Squid memory usage.
maximum_object_size_in_memory 128 KB
The "mem" report shows me the following top-4 lines
Pool Allocated
In Use
(#) (KB)
(#) (KB)
32K Buffer 20955 670560 20955
670560
16K Buffer 21380 342080 21378
342048
8K Buffer 11547 92376 11546
92368
Short Strings 1851173 65081 1851173
65081
What are the generic 32/16/8K buffers and what is the next thing I can
investigate to figure it out? I'm in way over my head and not sure where to
go next.
See the cache manager "memory" report to get details of what allocation
types the memPools are tracking. Read this thread from the beginning to
see how Mike debugged the ident usage and compare it to your repliation
of what he did.
Amos
