patch for HEAD and SAMBA_2_2
t.d.lee at durham.ac.uk
Wed Sep 26 04:03:02 GMT 2001
On Tue, 25 Sep 2001, Scott Moomaw wrote:
> When the load average begins to climb, have you looked at the number of
> smbd processes that exist? I have seen several occasions where the system
> under heavy load starts spawning new smbd processes at a high rate. It is
> like some timeout is happening so a new process is being spawned. Given a
> short amount of time in this condition, the server can spawn 1000s of new
> smbd processes exacerbating this problem to the point that killing samba
> is the only solution.
Indeed. We do seem to have an increased number of smbds when this
happens. But the load is so high that "top" grinds to a halt and any
other analysis is also nigh impossible. Oh, and the users are
understandably applying some pressure to restore service...
I suspect something like the following:
1. For whatever reason (e.g.network) a pause has occured;
2. Client PCs think they have timed out and try to establish a
3. These appear as extra smbds; i.e. now 2(+) smbds per PC instead of 1;
4. Perhaps these multiples have contention on tdbs.
Note that after stopping and starting samba, there is a similar rapid rise
of smbds, but the load impact of this is small, presumably as there is
only one per PC, and presumably therefore no "competition" for a given tdb
More information about the samba-technical