Subject | running two instances of the 2.1rc2 superserver on a 32bit linux hell experience |
---|---|
Author | woodsmailbox |
Post date | 2008-06-11T13:47:53Z |
I was afraid to post this directly to the bugtracker list because I
don't know anymore what's a bug and what's just normal inconsistent
behavior for eternal punishment of the DBA. So I'll just flush out my
experience here.
fbmgr.bin: trying to stop the server through it gives "Invalid
clumplet buffer structure: buffer end of clumplet - no length
component. cannot attach to server". it could just as well say
"010100100100 beep beep beeeeep beep beep". "clumplet" isn't an
English word.
fbmgr.bin: trying to start two server instances through it fails
(error: server already running), even after setting/exporting
$FIREBIRD to a different dir, configured firebird to listen to a
different ip address and giving a different pid file on the command line.
I gave up on fbmgr and now starting fbserver directly, managing its
pid file and all from a bash script, etc. it seems to work, although I
admit I didn't tested it that much yet.
fbguard: won't work with two server instances. gave up on that one too.
fbserver: the only program in the suite that takes server's root dir
as a command line parameter. other programs like fbmgr, fbguard, gbak,
etc. only watch for the FIREBIRD env. var. looks like a small thing to
rant about but this torture of inconsistency makes scripts hard to follow.
firebird.conf: it's nice that we have RemoteBindAddress, but why does
it have to listen on localhost too unconditionally? what happens if I
run two servers configured with different RemoteBindAddress? which
server opens the localhost listening socket?
nbackup: will accept localhost:db-alias but not any-hostname:db-alias
for the -B and -R switch. It *will* accept a full path instead of
alias for the -R switch (and will work without a running server) but
not for the -B switch. Don't tell me I didn't RTFM, because it doesn't
even matter whether this kind of behavior is documented anywhere or not.
I think these problems happen/don't get talked about because of the
limited user base of such configuration (two server instances on a
production linux server + automated backup/restore scripts, etc.). I
myself for years only had firebird on my windows desktop machine,
never compiled it, never did any backups/restores (that's what most
users have, I think).
don't know anymore what's a bug and what's just normal inconsistent
behavior for eternal punishment of the DBA. So I'll just flush out my
experience here.
fbmgr.bin: trying to stop the server through it gives "Invalid
clumplet buffer structure: buffer end of clumplet - no length
component. cannot attach to server". it could just as well say
"010100100100 beep beep beeeeep beep beep". "clumplet" isn't an
English word.
fbmgr.bin: trying to start two server instances through it fails
(error: server already running), even after setting/exporting
$FIREBIRD to a different dir, configured firebird to listen to a
different ip address and giving a different pid file on the command line.
I gave up on fbmgr and now starting fbserver directly, managing its
pid file and all from a bash script, etc. it seems to work, although I
admit I didn't tested it that much yet.
fbguard: won't work with two server instances. gave up on that one too.
fbserver: the only program in the suite that takes server's root dir
as a command line parameter. other programs like fbmgr, fbguard, gbak,
etc. only watch for the FIREBIRD env. var. looks like a small thing to
rant about but this torture of inconsistency makes scripts hard to follow.
firebird.conf: it's nice that we have RemoteBindAddress, but why does
it have to listen on localhost too unconditionally? what happens if I
run two servers configured with different RemoteBindAddress? which
server opens the localhost listening socket?
nbackup: will accept localhost:db-alias but not any-hostname:db-alias
for the -B and -R switch. It *will* accept a full path instead of
alias for the -R switch (and will work without a running server) but
not for the -B switch. Don't tell me I didn't RTFM, because it doesn't
even matter whether this kind of behavior is documented anywhere or not.
I think these problems happen/don't get talked about because of the
limited user base of such configuration (two server instances on a
production linux server + automated backup/restore scripts, etc.). I
myself for years only had firebird on my windows desktop machine,
never compiled it, never did any backups/restores (that's what most
users have, I think).