Search squid archive

Re: Scaling concurrent TCP sessions beyond ephemeral port range

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 




On Thu, Sep 8, 2022 at 8:31 PM Alex Rousskov <rousskov@xxxxxxxxxxxxxxxxxxxxxxx> wrote:
On 9/8/22 19:41, Praveen Ponakanti wrote:
>   * We have a large number of workers (30) to help with handling a
>     high RPS. However, TCP session reuse does not seem to be optimal
>     even with server_persistent_connections enabled as a new outbound
>     session would have to be opened up if the request is proxied by a
>     kid worker that doesn’t already have a connection to that
>     destination. Is there something that can be done to improve this
>     with later versions of squid? Would be glad to help out if anyone
>     has some suggestions.

If your only concern is TCP, and the number of servers is large, then it
would be possible to share open Squid-server connections among workers
by adding code that would exchange open TCP socket descriptors using UDS
messages, but I doubt it is worth doing (a lot of complexity but not
enough gain). There may also be some advanced/modern kernel tricks that
we can teach Squid to use for sharing connections, but, again, I doubt
the complexity would be worth the benefits from such reuse.

Agree, it might not make sense to increase the complexity with sharing socket among the workers. Was thinking more on the lines of a hashmap that the coordinator could use to pick workers that already have a TCP connection to the destination being requested, instead of having the workers themselves share connection details. This might introduce significant complexity as well, so please ignore if there are better solutions when large counts of workers are in use.
 

If most TCP servers are known a priori, and there are few of them, then
cache_peer standby=N feature for them might be useful.


The TCP destinations are not always known beforehand and can run into the thousands. However only the top 4-5 destinations have a large number of concurrent sessions. Which are each limited to the ip_local_port_range size in concurrent sessions now after the enhancement to add the IP_BIND_ADDRESS_NO_PORT flag. BTW, we do not run caching on squid or have peer caches in our deployment.  

Most of the TCP connections are for HTTPS reqs, w/o TLS termination at the squid. Does squid currently support a TLS session cache ?

Thanks
Praveen
 
If you are dealing with TLS sessions as well, then we should add a
shared memory TLS session cache that all workers can tap into.


Cheers,

Alex.

> On Tue, Jun 21, 2022 at 2:11 PM Alex Rousskov wrote:
>
>     On 6/19/22 12:48, Praveen Ponakanti wrote:
>
>      > What is the process to have this code patch upstreamed for future
>     squid
>      > versions?
>
>     In short, just post a quality pull request on GitHub (or find somebody
>     who can guide your code towards official acceptance for you). For
>     details, please see https://wiki.squid-cache.org/MergeProcedure
>     <https://wiki.squid-cache.org/MergeProcedure>
>
>
>     Thank you,
>
>     Alex.
>
>
>      > On Fri, May 20, 2022 at 9:31 PM Amos Jeffries
>     <squid3@xxxxxxxxxxxxx <mailto:squid3@xxxxxxxxxxxxx>
>      > wrote:
>      >
>      >     On 20/05/22 19:44, Praveen Ponakanti wrote:
>      >      > Hi Alex,
>      >      >
>      >      > Thanks for going through several steps to help mitigate
>     src port
>      >      > exhaustion. We are looking to achieve 400-500% more
>      >      > concurrent connections if we could :) as there is a
>      >     significant buffer
>      >      > on the available CPU.
>      >
>      >     Then you require at least 4, maybe 5, IP addresses to handle
>     that many
>      >     concurrent connections with Squid.
>      >
>      >
>      > We would like to investigate going beyond the ephemeral port
>     range for
>      > some specific destination IP:PORT addresses. For that it appears
>     squid
>      > does not round-robin requests if we use multiple
>     tcp_outgoing_addresses.
>      > We could use ACL’s to pick a different outbound IP based on the
>     clients
>      > source IP, however that is not very ideal in our environment as our
>      > clients aren’t always equally split by subnet. However, if we could
>      > split by the client’s source port that might help achieve this. For
>      > example something like:
>      >
>      >
>      > acl pool1 clientport 0-32768
>      >
>      > acl pool2 clientport 32769-65536
>      >
>      >
>      > tcp_outgoing_address 10.1.0.1 pool1
>      >
>      > tcp_outgoing_address 10.1.0.2 pool2
>      >
>      >
>      > Squid's ACLs currently do not allow filtering by the client's source
>      > port. We could look into a separate patch to add this
>     functionality to
>      > squid’s ACL code if that makes sense. Or is there a better way to
>      > achieve this?
>      >
>      >
>      > Thanks
>      >
>      > Praveen
>      >
>      >
>      >      > The option to use multiple tcp_outoing_addresses appears to be
>      >     promising
>      >      > along with some tweaks to the TCP timeouts. I guess we
>     could use
>      >     ACLs to
>      >      > pick a different outbound IP based on the requesting client's
>      >     prefix. We
>      >      > had not considered that option as the ephemeral ports were
>     no longer
>      >      > available to other applications when squid uses most of
>     them with a
>      >      > single outbound IP configured. We are also looking to
>     modify the
>      >     code to
>      >      > use the IP_BIND_ADDRESS_NO_PORT sockopt as that could help
>     delay
>      >     port
>      >      > assignment with the bind() call on the outbound TCP
>     sessions (to
>      >      > hopefully allow access to the 4-tuple on the socket).
>      >
>      >     Patches welcome.
>      >
>      >     However, please be aware that use of the 4-tuple is often no
>     different
>      >     from the 3-tuple since the dst-port is typically identical
>     for all
>      >     outgoing traffic to a given dst-IP.
>      >
>      >
>      >     Cheers
>      >     Amos
>      >     _______________________________________________
>      >     squid-users mailing list
>      > squid-users@xxxxxxxxxxxxxxxxxxxxx
>     <mailto:squid-users@xxxxxxxxxxxxxxxxxxxxx>
>      >     <mailto:squid-users@xxxxxxxxxxxxxxxxxxxxx
>     <mailto:squid-users@xxxxxxxxxxxxxxxxxxxxx>>
>      > http://lists.squid-cache.org/listinfo/squid-users
>     <http://lists.squid-cache.org/listinfo/squid-users>
>      >     <http://lists.squid-cache.org/listinfo/squid-users
>     <http://lists.squid-cache.org/listinfo/squid-users>>
>      >
>      >
>      > _______________________________________________
>      > squid-users mailing list
>      > squid-users@xxxxxxxxxxxxxxxxxxxxx
>     <mailto:squid-users@xxxxxxxxxxxxxxxxxxxxx>
>      > http://lists.squid-cache.org/listinfo/squid-users
>     <http://lists.squid-cache.org/listinfo/squid-users>
>
>     _______________________________________________
>     squid-users mailing list
>     squid-users@xxxxxxxxxxxxxxxxxxxxx
>     <mailto:squid-users@xxxxxxxxxxxxxxxxxxxxx>
>     http://lists.squid-cache.org/listinfo/squid-users
>     <http://lists.squid-cache.org/listinfo/squid-users>
>

_______________________________________________
squid-users mailing list
squid-users@xxxxxxxxxxxxxxxxxxxxx
http://lists.squid-cache.org/listinfo/squid-users

[Index of Archives]     [Linux Audio Users]     [Samba]     [Big List of Linux Books]     [Linux USB]     [Yosemite News]

  Powered by Linux