At 20:41 -0700 4/25/02, Clay Loveless wrote:
>Thanks for your input into this problem!
>1) That's an interesting point about the HFS+ filesystem -- I've read that a
>G4 won't boot unless the startup volume is formatted as HFS+ (so sayeth the
>10.1 Server Update "README FIRST.pdf") ... So, just for consistency's sake,
>both drives in my server are formatted for HFS+. But, I assume HFS+ is not
>the most common filesystem MySQL is installed upon ... So who knows? That
>could be an issue.
>2) I don't actually actually query an explicit filename reference -- I'm
>doing a standard mysql_connect/mysql_pconnect through PHP to a the hostname
>of our MySQL server. We have a two-webserver and one MySQL server setup,
>with all three machines running identical hardware and OS configurations,
>all in the same rack on a 100BaseT network (so there shouldn't be any real
>network latency issues) ... But connection is being made over a network, not
>to the "localhost".
>But, that did get me to thinking about mysql_connect vs mysql_pconnect ...
>When I moved all our stuff from FreeBSD boxes to Mac OS X Server, all of our
>database calls through PHP were made via mysql_pconnect. I've had some other
>oddities (rapid memory consumption by only ONE of the OS X web servers)
>which popped up during that move, and I've been experimenting with switching
>back to mysql_connect().
It's worth a shot. Persistent PHP connections have more benefit with
other database engines for which connection establishment is more of
a heavyweight operation. For MySQL, the cost is very low -- and reduces
the likelihood of getting problems from having zillions of connections
>Since a socket connection is considered a file descriptor, I wonder if
>something funny is going on that causes a persistent connection to "go bad",
>thus generating "bad file descriptor" errors?
>If that were true, I could switch to using solely mysql_connect() calls,
>since I believe that opens a new connection, uses it, and closes it ...
>Leaving no real opportunity for that particular connection's file descriptor
>to go bad.
>I'm definitely groping around on this one myself ... It's easy enough to
>switch to using only mysql_connect(), and I can do that and see how things
>go for awhile ... But it's one of those things where it's a "wait and see
>with fingers crossed" remedy.
>Barry, Paul, Mark, Erik ... Anyone else ... Opinion on that theory?
>Thanks again, folks -- I appreciate the group brainstorm.
>(PS -- I'd email Marc Liyanage, but I can only imagine how many people are
>bugging him already! I'm using his latest binary build at the moment.)