On Mon, 25 Sep 2000, Sergey Verzunov wrote:
> We've just started testing using polygraph and facing some problems.
> Could somebody, please, help us with the following.
>
> Runtime
> 1. polyclt error not documented: "Unsupported HTTP response code..."
> Our investigation have shown that this "unsupported" code is 502 Bad
> gateway. Is this correct that polyclt does not support 502 code?
Polyclt understands 200 and 304 codes only. There was simply no reason
to support other types of responses.
> 2. Further investigation of problem 1. have shown, that polysrv in some
> cases sends with FIN (closes connection)
> packet after accepting a connection and GET request from the cache.
> What is the reason for such a behaviour?
I can think of the two primary reasons:
- Polysrv detects an error (the error will then be reported).
Most errors do not cause the connection to be closed, but
some do
- Polysrv closes an idle persistent connection _before_
reading the request; keep in mind that TCP stack may read
the request into buffers before the server knows anything
about that request. This race condition is a "normal" HTTP
behavior in the presence of persistent connections. If you
disable pconns and the error goes away, then my guess is
correct.
> 3. Sometimes Error: "Too many open files" comes during a test.
> How many sockets and open files should be configured in a FreeBSD 3.3
> system to avoid this error?
> We put number 64K.
64K should be sufficient. I assume you have tuned other OS resources to
match your request rates. You may want to read the "tips" page on
Polygraph Web site or, better, get an up-to-date snapshot from
measurement-factory.com. "Too many open files" might indicate that OS
lacks some other resources.
> 4. How do we remove server xaction delay: xact_think does not work in 2.x?
Think time is specified on a per-server basis. For example, see PGL
Server configurations in PolyMix-2 and PolyMix-3 workload files.
> Results
> 1. After a test we have a log file. Here is part of "lr file_name.log"
> outout:
>
> ----------------------------------------------------------------------------
> 969837155.041544| ./CltXact.cc:399: error: 127/127 (272) clocks out of sync
> 969837155.041544| 969837233.000000 - 969837155.041544 = 1.30min
BTW, you probably want to fix this error.
>
> But. label_results, or lx programs return the following errors:
>
> ----------------------------------------------------------------------------
> ./label_results: creating: `/tmp/polyrep/test_ws/clt.All.lx'
> no matching phases found
> ./label_results: `/usr/local/polygraph/bin/lx --out
> /tmp/polyrep/test_ws/clt.All.lx
> .//clt_ws.log' failed; probable cause: No such file or directory
> ----------------------------------------------------------------------------
>
> As you can see test log file has all phases in place, test log file itself
> also in place. What can be the reason for these errors?
Does "lx clt_ws.log" work? What does "lr clt_ws.log | tail" return?
Please note that if polyclt is killed before the end of the test, phase
statistics is not logged. Lr extracts console output from binary logs;
just because you can see that output does not mean that the phase stats
got logged.
Alex.
This archive was generated by hypermail 2b29 : Tue Jul 10 2001 - 12:00:15 MDT