Home / server / Questions / 239910
Accepted
Matt Beckman
Asked: 2011-02-25 11:51:49 +0800 CST

Using Squid cache_peer for performance in an Apache cluster

  • 772

I have 3 nodes in a web server cluster. Each node is running Squid on the LAN interface and Apache on the localhost interface. Squid caches properly for each server individually, but I don't believe I setup cache peers correctly.

cache_peer 127.0.0.1 parent 80 0 no-query originserver name=Server1
cache_peer Server2 sibling 80 3130 proxy-only
cache_peer Server3 sibling 80 3130 proxy-only
cache_peer_access Server1 allow MYDOMAIN

Each server has similar cache_peer settings except each is its own parent and uses the other two servers as siblings. My goal is to improve hit rate as much as possible, but I don't know if I'm understanding the cache peer logic correctly.

I want Squid to try to find http://www.example.com/mypage/ in it's local cache, and if it's not query the siblings until it finds it. If it can't find it, THEN it should hit Apache.

Am I doing this right?

apache-2.2 squid reverse-proxy

2 Answers

  1. Best Answer

    Squid tries to query each peer via ICP, measures response times and then decides which parent to use. There is common problems:

    • your peers doesn't respond on ICP queries and squid marks them as DEAD
    • your peers doesn't support cache digest
    • you are serving not cacheable content and squid always connects directly
    • your acl rules doesn't allows use other peers

    Answer is in your access logs. so try this and give us an example: tail -f /var/log/squid/access.log

    • 1

  2. So in this example, do you have squid listening to your external IP on port 80, and apache bound to localhost:80 on each server?

    One thing to check is to make sure you have icp_port 3130 if you're running squid 3.0 or above; in 2.x versions 3130 it was the default but in 3.x it's now 0 (disabled).

    • 0