Jeff Peng wrote: > Amos Jeffries 写道: >> Jeff Peng wrote: >>> Hi, >>> >>> I set up two groups of rules for the same accelerated domain name, but >>> the selection arithmetic and original servers are different. >>> Why squid doesn't complain about this setting? Which rule will squid >>> choose when accept that a request? Thanks. >> Squid doesn't complain because its a possible configuration, with peer >> failover. >> >> What it does is when looking to find a peer, checks each possible >> selection method. First that produces usable peers wins. >> >> In this case sourcehash is more preferred than round-robin, so... >> One of C and D will be picked based on the source hash. >> If C and D are (both?) down then it will failover to try A and B based >> on the round-robin method. > > > Thanks Amos. > In this case, why squid prefer sourcehash group more than round-bin > group? does sourcehash has higher PRI than round-bin? Thanks again. > The preferences are really hard-coded order of testing each method. I think we've sorted them by efficiency or stability or something. The new ones I've seen go in, certainly went into specific places middle of the list. FWIW the order I read in the code is: (always direct? -> DIRECT) (filter out any peers whose ACL deny their use) sibling or parent with object (digests) closest parent peer (netdb RTT measured) default parent userhash parent srchash parent carp parent round-robin parent weighted-round-robin parent first parent any parent DIRECT Amos -- Please use Squid 2.7.STABLE4 or 3.0.STABLE9
