ACE Leastconn question

[dknov]

Hi,

I had just moved one server farm from round-robin to leastconns with slowstart of 300 second and no new rservers had been added (or failed), they are all the same as before the change.

What I see is that one rserver gets much more hits than other and one of them is practically idle. I know that CSCso93479 states that current connections count in "show serverfarm" is inaccurate, but I cannot understand such a difference....

Is total connections counter bugged as well?

----------connections-----------

real weight state current total failures

---+---------------------+------+------------+----------+----------+---------

rserver: prmesapp11

10.16.127.17:0 8 OPERATIONAL 23 6 0

rserver: prmesapp12

10.16.127.18:0 8 OPERATIONAL 44 187 0

rserver: prmesapp13

10.16.127.19:0 8 OPERATIONAL 31 43 0

rserver: prmesapp14

10.16.127.20:0 8 OPERATIONAL 27 62 0

Or am i missing something about leastconns predictor?

Thanks a lot!

David

[Gilles Dufour]

do you have stickyness ?

G.

[dknov]

Hi,

Nope. I was trying to see what I can do about this, so I removed the leastconn (reverting back to round-robin) then configured leastconn back, but without slow-start parameter. What I immediately noticed is that servers started to be hit in a more equal manner, which is what I expected. I then reapplied the leastconn command, but with slow-start parameter and it would seem that session distribution was as expected. I assume that maybe removing and reapplying leastconn command did the trick, or maybe slow-start parameter was somehow misbehaving when I first applied it....

Now what I noticed is there are some sessions under failure column of "show serverfarm" output and I don't believe I had those before I switched to leastconn. The number is very low, like 5 failed versus 30,000 total, but still I was wondering if there is anything different with leastconn from round-robin that would cause some of the sessions to fail ?

Thanks!

David

[Gilles Dufour]

David,

we have a few slowstart issue where the internal data structure could get corrupted preventing some servers to receive traffic.

You may have hit such a problem.

CSCsx80991: ACE: Leastcon negative weight after adding new rservers

Usually, removing/adding the predictor does fix the problem.

Next time you see the issue, you should open a service request to receive guidance on what information to collect.

The procedure is not as easy as show ....

Regarding the failure, it is independent of the predictor.

We count a failure when the connection can't be established.

Reset from server or no response.

Gilles.