> On Dec 17, 2016, at 23:13, Assaf Gordon <assafgor...@gmail.com> wrote:
>
>> On Dec 17, 2016, at 16:48, Bob Proulx <b...@proulx.com> wrote:
>>
>> [...] Yesterday we started to see some spikes which hit the process
>> limits of 150 for a couple of hours.
>
> This does seem to coincide with the cvs switch to the new server.
I have verified (with strace,ltrace) that the new cvs binary running in
'pserver' mode as user 'nobody' does call "getpwnam(3)" on user 'nobody' which
triggers an database query due to libnss-mysql.
===
root@vcs0:/tmp# grep -A 4 getpwnam
cvs.ltrace-100.37.92.116-2016-12-17-234502.log
13870 getpwnam("nobody") = 0x7f18a8399240
root@vcs0:/tmp# grep -C 4 getpwnam
cvs.strace-100.37.92.116-2016-12-17-234334.log
13325 futex(0x7fd8abf811c0, FUTEX_WAKE_PRIVATE, 2147483647) = 0
13325 open("/etc/libnss-mysql.cfg", O_RDONLY) = 3
13325 fstat(3, {st_mode=S_IFREG|0644, st_size=6423, ...}) = 0
13325 mmap(NULL, 4096, PROT_READ|PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0)
= 0x7fd8ae7de000
13325 read(3, "getpwnam SELECT user_name, 'x"..., 4096) = 4096
13325 read(3, "ND groups.gidNumber >= 1000 \\\n# "..., 4096) = 2327
13325 read(3, "", 4096) = 0
13325 close(3) = 0
13325 munmap(0x7fd8ae7de000, 4096) = 0
===
I wonder if there's a way to tell libnss-mysql to skip the query for local
users if they are in /etc/passwd ?
Or should we look into modifying the cvs binary (perhaps there's a simply
runtime option).
I'll look into it.
- assaf
