Hello,
I suspected I might be experiencing the looping problem Theo mentioned, so I
updated to the CVS version of mod_backhand recently.
After the upgrade I started getting some odd behaviour with lots of error
messages in the error log. I disabled the connection pools and it seems to
be fine now. Is this a known problem? The log entries are below.
---- logfile ----
[Thu Nov 7 14:30:55 2002] [error] mod_backhand: Check 6 -- headers(-1)
[Thu Nov 7 14:30:55 2002] [warn] (4)Interrupted system call: connect()
timed out
[Thu Nov 7 14:30:55 2002] [warn] (4)Interrupted system call: connect()
timed out
[Thu Nov 7 14:30:55 2002] [warn] (4)Interrupted system call: connect()
timed out
[Thu Nov 7 14:30:55 2002] [warn] (4)Interrupted system call: connect()
timed out
[Thu Nov 7 14:30:55 2002] [warn] (4)Interrupted system call: connect()
timed out
[Thu Nov 7 14:30:55 2002] [warn] (4)Interrupted system call: connect()
timed out
=== snip 60 lines ===
[Thu Nov 7 14:30:55 2002] [warn] (4)Interrupted system call: connect()
timed out
[Thu Nov 7 14:30:55 2002] [warn] (4)Interrupted system call: connect()
timed out
[Thu Nov 7 14:30:55 2002] [warn] (4)Interrupted system call: connect()
timed out
[Thu Nov 7 14:30:55 2002] [warn] (4)Interrupted system call: connect()
timed out
[Thu Nov 7 14:30:55 2002] [warn] (4)Interrupted system call: connect()
timed out
[Thu Nov 7 14:30:55 2002] [warn] (4)Interrupted system call: connect()
timed out
[Thu Nov 7 14:30:55 2002] [warn] (4)Interrupted system call: connect()
timed out
[Thu Nov 7 14:30:55 2002] [error] (9)Bad file descriptor: mod_backhand:
MBCSP error (recv file descriptor)
Anyone else seeing this?
A couple more questions, how does the addPrediction work? Do you have to
give it any parameters or does it just adjust the stats of the machine that
is going to be the destination of the request. This only affects the stats
being kept by the proxying machine and not cluster wide. Correct?
Why does Backhand not use spread for group communication? If the proxying
host(s) were to announce that a request were being sent to a particular
backend machine, then every memeber of the group would get it and could
adjust its local stats accordingly. Perhaps the load imposed by all these
spread messages would nullify the benefit of having higher resolution stats
(than the 1 second intervals we have now).
Thanks for any info, and thanks for the great program guys!
Dennis
I suspected I might be experiencing the looping problem Theo mentioned, so I
updated to the CVS version of mod_backhand recently.
After the upgrade I started getting some odd behaviour with lots of error
messages in the error log. I disabled the connection pools and it seems to
be fine now. Is this a known problem? The log entries are below.
---- logfile ----
[Thu Nov 7 14:30:55 2002] [error] mod_backhand: Check 6 -- headers(-1)
[Thu Nov 7 14:30:55 2002] [warn] (4)Interrupted system call: connect()
timed out
[Thu Nov 7 14:30:55 2002] [warn] (4)Interrupted system call: connect()
timed out
[Thu Nov 7 14:30:55 2002] [warn] (4)Interrupted system call: connect()
timed out
[Thu Nov 7 14:30:55 2002] [warn] (4)Interrupted system call: connect()
timed out
[Thu Nov 7 14:30:55 2002] [warn] (4)Interrupted system call: connect()
timed out
[Thu Nov 7 14:30:55 2002] [warn] (4)Interrupted system call: connect()
timed out
=== snip 60 lines ===
[Thu Nov 7 14:30:55 2002] [warn] (4)Interrupted system call: connect()
timed out
[Thu Nov 7 14:30:55 2002] [warn] (4)Interrupted system call: connect()
timed out
[Thu Nov 7 14:30:55 2002] [warn] (4)Interrupted system call: connect()
timed out
[Thu Nov 7 14:30:55 2002] [warn] (4)Interrupted system call: connect()
timed out
[Thu Nov 7 14:30:55 2002] [warn] (4)Interrupted system call: connect()
timed out
[Thu Nov 7 14:30:55 2002] [warn] (4)Interrupted system call: connect()
timed out
[Thu Nov 7 14:30:55 2002] [warn] (4)Interrupted system call: connect()
timed out
[Thu Nov 7 14:30:55 2002] [error] (9)Bad file descriptor: mod_backhand:
MBCSP error (recv file descriptor)
Anyone else seeing this?
A couple more questions, how does the addPrediction work? Do you have to
give it any parameters or does it just adjust the stats of the machine that
is going to be the destination of the request. This only affects the stats
being kept by the proxying machine and not cluster wide. Correct?
Why does Backhand not use spread for group communication? If the proxying
host(s) were to announce that a request were being sent to a particular
backend machine, then every memeber of the group would get it and could
adjust its local stats accordingly. Perhaps the load imposed by all these
spread messages would nullify the benefit of having higher resolution stats
(than the 1 second intervals we have now).
Thanks for any info, and thanks for the great program guys!
Dennis