[squid-users] A lot of open rewriter heplers and are hanging! Squid 3.5

[Amos Jeffries]

On 5/05/2015 3:35 a.m., Ulises Nicolini wrote:
> Hello Amos and Eliezer,
> 
> Regarding this:
> 
>> My guess is that its been patched to cope with the action code and
>> kv-pair syntax. But not concurrency enabled. Which is mandatory on the
>> Store-ID interface.
> 
> You are right about the kv-pair syntax patch, that's exactly it. But
> about concurrency, we're running Squid as a unique process. Do we need
> to modify jesred in order to handle Store-ID interface? The only
> modification we did to jesred is that we changed the output string that
> it gives back to Squid.
> 

The Store-ID interface requires concurreny support in the helpers it uses.

For your jesred it would mean receiving the channel-ID value at the
start of each request line from Squid and echoing it back as a prefix on
the answer lines.


> 
> About this:
> 
>> You really sure 20 children is enough for 1200 clients? Also whenever
>> bypass on?
> 
> Is there any formula to calculate how many children should we use based
> on the amount of requests per second? Or is it just try and error?
> 

There is, but you need to know the helper req/sec capacity to alculate it.

The easy way with current Squid is to set a high children number, with a
lower start=N and small idle=N parameters. Squid starts what helpers are
needed dynamically.


> 
> We' re getting the impresion that jesred processes are waiting for some
> kind of signal to die once they're no longer necessary, ergo they just
> stay there, doing nothing.
> 

If a helper dont have concurrency but Squid is expecting it to, the
first request may "work" but a second one will definitely hang.

Amos