| Subject | Re: [IB-Java] Re: Dying connections with interclient |
|---|---|
| Author | Ola Samuelson |
| Post date | 2001-04-15T20:27:05Z |
Hi!
are lost which means new ones has to be created. But creating
new ones while old ones are "dying" takes "forever". I goes real fast to create
like
10 connections the first time.
I have a connection pool which tries to revive the connections but
even with this the connections that I am reviving(restarting)
are slow. Also, Interserver processes related to dead connections
are not returning memory.
So after a while I have like a few megs avail even with no heavy load.
I have 512M in the system and after starting all processes the first
time some 35megs are occupied.
So the server is spiraling down :-(
It does not matter if they are constantly used or not.(If i remember correctly)
I only use get meta data when starting the connection pool to get number
of allowed connections.
I have not got a clue what's going on.
1. Connections are dying after a few minutes
2." connection.isclosed()" does not really tell me if it is closed or not. I
worked a bit better when I did dummy reads to check
the state of the connection. But this cost to much time.
3. I am using a pool which many have tried and are extremely happy with. AND I
have tried others as well. So my guess it's not that.
4. Even with a pool full of what is supposed to be available connections - at
time of use the connection is found dead and
a new one needs to be created. Creation of new ones at this point in time takes
30 secs sometimes.
5. there has been some spurious exceptions indicating inability of IC to
resolve database url
Could it be a glibc/jdk/interclient combination problem?
I think I have tried them all by now.
Right now I am on:
glibc 2.1.3-22(original was 2.1.3-21)
JDK 1.2.2_007(tried all 1.2's from blackdown and suns 1.3.0_02)
Apache Jserv 1.1.2 (got from CVS)
Apache JSSI 1.1.2
Intel PIII SMP machine based on RH6.1 with all updates
Same behavior even with the db locally so I guess it's not a network problem.
I have tried setting various tcp parameters with sysctl and it did seem to
improve things
a bit since it kills connections in fin_wait state sooner. And the interserver
processes
related to dead connections are reported to be in state fin_wait or time_wait.
Detail: NIC is Intel EEPRO 100 with Donald Beckers driver from february I
think.
I did compile interserver from cvs a few weeks ago and used it in conjuction
with
interclient.jar supplied by Joe Shevland(interbase2000.org).
No difference .........
Like I said I have tried the db locally no change but the current one
is behind a firewall with filtering and portforwarding allowing 3050/tcp
to pass between the machines. The server is running IB 6.01CS on linux
smp 2.2.18 kernel. It all worked better when we ran IB 5.6 - on the server
but we need the 6.01 features so...... Any server side configs that enables
a remote interserver to work better?
I find it confirmed after going through and following IC newsgroups and
mailinglists for more than 18 months that there are at least three problems
with all downloadable versions of interclient/interserver:
1. Dying connections - seen a lot of mail about that one
2. Interserver eating memory
3. Inablility to use more than ~20 connections under Linux.
True or false?
Sorry for my ramblings - ask away if I am being unclear!
Ideas or questions?
Many thanks in advance.
//OLAS
PS.
One thing I have noticed: I get a "SIGPIPE" error thrown by
the servlet engine if I transfer more than a few K's from over the connection.
And this ONLY happens when accessed with Netscape over the web.
Not with IE. I don't know if it's related in any way.
> hours (overnight, 8 hours or so) of disuse the connections break, soIn my case it's a matter of a few minutes and after that the connections
are lost which means new ones has to be created. But creating
new ones while old ones are "dying" takes "forever". I goes real fast to create
like
10 connections the first time.
I have a connection pool which tries to revive the connections but
even with this the connections that I am reviving(restarting)
are slow. Also, Interserver processes related to dead connections
are not returning memory.
So after a while I have like a few megs avail even with no heavy load.
I have 512M in the system and after starting all processes the first
time some 35megs are occupied.
So the server is spiraling down :-(
It does not matter if they are constantly used or not.(If i remember correctly)
I only use get meta data when starting the connection pool to get number
of allowed connections.
I have not got a clue what's going on.
1. Connections are dying after a few minutes
2." connection.isclosed()" does not really tell me if it is closed or not. I
worked a bit better when I did dummy reads to check
the state of the connection. But this cost to much time.
3. I am using a pool which many have tried and are extremely happy with. AND I
have tried others as well. So my guess it's not that.
4. Even with a pool full of what is supposed to be available connections - at
time of use the connection is found dead and
a new one needs to be created. Creation of new ones at this point in time takes
30 secs sometimes.
5. there has been some spurious exceptions indicating inability of IC to
resolve database url
Could it be a glibc/jdk/interclient combination problem?
I think I have tried them all by now.
Right now I am on:
glibc 2.1.3-22(original was 2.1.3-21)
JDK 1.2.2_007(tried all 1.2's from blackdown and suns 1.3.0_02)
Apache Jserv 1.1.2 (got from CVS)
Apache JSSI 1.1.2
Intel PIII SMP machine based on RH6.1 with all updates
Same behavior even with the db locally so I guess it's not a network problem.
I have tried setting various tcp parameters with sysctl and it did seem to
improve things
a bit since it kills connections in fin_wait state sooner. And the interserver
processes
related to dead connections are reported to be in state fin_wait or time_wait.
Detail: NIC is Intel EEPRO 100 with Donald Beckers driver from february I
think.
I did compile interserver from cvs a few weeks ago and used it in conjuction
with
interclient.jar supplied by Joe Shevland(interbase2000.org).
No difference .........
Like I said I have tried the db locally no change but the current one
is behind a firewall with filtering and portforwarding allowing 3050/tcp
to pass between the machines. The server is running IB 6.01CS on linux
smp 2.2.18 kernel. It all worked better when we ran IB 5.6 - on the server
but we need the 6.01 features so...... Any server side configs that enables
a remote interserver to work better?
I find it confirmed after going through and following IC newsgroups and
mailinglists for more than 18 months that there are at least three problems
with all downloadable versions of interclient/interserver:
1. Dying connections - seen a lot of mail about that one
2. Interserver eating memory
3. Inablility to use more than ~20 connections under Linux.
True or false?
Sorry for my ramblings - ask away if I am being unclear!
Ideas or questions?
Many thanks in advance.
//OLAS
PS.
One thing I have noticed: I get a "SIGPIPE" error thrown by
the servlet engine if I transfer more than a few K's from over the connection.
And this ONLY happens when accessed with Netscape over the web.
Not with IE. I don't know if it's related in any way.
>
> although I encourage you to try my changes I doubt they will completely fix
> the problem. I haven't put up a compiled version, I was waiting for
> someone to try it out on windows, and got involved in some other stuff.
> You can check out the 2.0 source and find the ant build script in the
> firebird directory. Let me know if you have any problems.
>
> Can you be more specific about what is happening? How long before the
> connections break? Are you using them all the time or are there long
> periods of doing nothing? The problems I know about happen after long
> periods of disuse. The problems I fixed were primarily that the
> DatabaseMetaData.getTables(...) call immediately crashed interserver. If
> you are using metadata methods it is very possible that my fix will help.
>
> Thanks
> David Jencks
>
> On 2001.04.13 16:46:28 -0400 Ola Samuelson wrote:
> > Hi!
> > Sure will try everything ..............
> >
> > Where did you say I can find the updated interclient?
> > TIA
> > //OLAS
> >
> > ft@... wrote:
> >
> > > I think there was a confirmed bug relating to this on Firebird
> > > Sourceforge site (Interbase shuts connection down without informing
> > > Interserver). However, the Firebird Interclient 2.00 has been updated
> > > extensively by David Jencks and can be compiled for Linux. There is a
> > > new Ant build.xml that makes it easy to build the jar and the
> > > Interserver executable. This version needs user feedback before
> > > general release so why not try it.
> > >
> > > Fred
> > >
> > > --- In IB-Java@y..., Ola Samuelson <ola@d...> wrote:
> > > > HI!
> > > > Tried them all ..............JDK's, operating systems, interclient
> > > kits.
> > > >
> > > > This happens:
> > > > Connections are dying or getting stale after just a few minutes.
> > > > Netstat says WAIT on connection between java/interserver.
> > > >
> > >
> > >
> > > To unsubscribe from this group, send an email to:
> > > IB-Java-unsubscribe@egroups.com
> > >
> > >
> > >
> > > Your use of Yahoo! Groups is subject to
> > http://docs.yahoo.com/info/terms/
> >
> >
> >
> > To unsubscribe from this group, send an email to:
> > IB-Java-unsubscribe@egroups.com
> >
> >
> >
> > Your use of Yahoo! Groups is subject to http://docs.yahoo.com/info/terms/
> >
> >
> >
> >
>
>
> To unsubscribe from this group, send an email to:
> IB-Java-unsubscribe@egroups.com
>
>
>
> Your use of Yahoo! Groups is subject to http://docs.yahoo.com/info/terms/