Putting Zope into a tailspin can kill a whole Linux Server
This is the second time I have done this and I just thought the list should know. Once in awhile (every six months or so) I send Zope a request that it just cannot handle. At that point (after waiting for a good five minutes) I cancel the request by stopping the browser. This has twice resulted in Zope spinning out of control (taking up more and more CPU) untill it kills the entire server box. How does it kill a Linux box you say? Well, it kills off Apache, kills off SSH (no more remote access), kills off FTP. They all just become unresponsive. Here are the two times it has happened: - Zope 2.3.2, on RH Linux 7.0 384 MB Ram, 650 MHz Pentium: I tried packing the database, stopped it and tried repacking it. - Zope 2.4.2 (same hardware): I tried creating a VERY LARGE Poll graph (10,000) and stopped it while it got going. I don't know if this is a bug or whatever, but it could be a problem. Zope can certainly be sent into a tailspin. BZ
BZ wrote:
This is the second time I have done this and I just thought the list should know.
Once in awhile (every six months or so) I send Zope a request that it just cannot handle. At that point (after waiting for a good five minutes) I cancel the request by stopping the browser. This has twice resulted in Zope spinning out of control (taking up more and more CPU) untill it kills the entire server box.
How does it kill a Linux box you say?
Well, it kills off Apache, kills off SSH (no more remote access), kills off FTP. They all just become unresponsive.
Here are the two times it has happened:
- Zope 2.3.2, on RH Linux 7.0 384 MB Ram, 650 MHz Pentium: I tried packing the database, stopped it and tried repacking it.
- Zope 2.4.2 (same hardware): I tried creating a VERY LARGE Poll graph (10,000) and stopped it while it got going.
I don't know if this is a bug or whatever, but it could be a problem. Zope can certainly be sent into a tailspin.
I am by no means an kernel expert (for whatever os), but it might be interesting to find out what exactly causes your box stalling. Are you sure that you aren't going OOM (out of memory) instead of "overloading" the cpu? What I mean is that the unresponsivenes(sp?) of your system might be caused by something other than zope's cpu-usage, i.e excessive memory usage. If you can repeat this, it might be worth using top/and or vmstat to watch what happens while zope goes to no man's land. cheers, oliver
You can also try the ForensicLogger product. ----- Original Message ----- From: "Oliver Bleutgen" <myzope@gmx.net> To: <zope@zope.org> Sent: Friday, November 02, 2001 12:30 PM Subject: Re: [Zope] Putting Zope into a tailspin can kill a whole Linux Server
BZ wrote:
This is the second time I have done this and I just thought the list should know.
Once in awhile (every six months or so) I send Zope a request that it just cannot handle. At that point (after waiting for a good five minutes) I cancel the request by stopping the browser. This has twice resulted in Zope spinning out of control (taking up more and more CPU) untill it kills the entire server box.
How does it kill a Linux box you say?
Well, it kills off Apache, kills off SSH (no more remote access), kills off FTP. They all just become unresponsive.
Here are the two times it has happened:
- Zope 2.3.2, on RH Linux 7.0 384 MB Ram, 650 MHz Pentium: I tried packing the database, stopped it and tried repacking it.
- Zope 2.4.2 (same hardware): I tried creating a VERY LARGE Poll graph (10,000) and stopped it while it got going.
I don't know if this is a bug or whatever, but it could be a problem. Zope can certainly be sent into a tailspin.
I am by no means an kernel expert (for whatever os), but it might be interesting to find out what exactly causes your box stalling. Are you sure that you aren't going OOM (out of memory) instead of "overloading" the cpu? What I mean is that the unresponsivenes(sp?) of your system might be caused by something other than zope's cpu-usage, i.e excessive memory usage. If you can repeat this, it might be worth using top/and or vmstat to watch what happens while zope goes to no man's land.
cheers, oliver
_______________________________________________ Zope maillist - Zope@zope.org http://lists.zope.org/mailman/listinfo/zope ** No cross posts or HTML encoding! ** (Related lists - http://lists.zope.org/mailman/listinfo/zope-announce http://lists.zope.org/mailman/listinfo/zope-dev )
BZ <bz@bwanazulia.com> wrote:
Once in awhile (every six months or so) I send Zope a request that it just cannot handle. At that point (after waiting for a good five minutes) I cancel the request by stopping the browser. This has twice resulted in Zope spinning out of control (taking up more and more CPU) untill it kills the entire server box.
[...]
- Zope 2.3.2, on RH Linux 7.0 384 MB Ram, 650 MHz Pentium: I tried packing the database, stopped it and tried repacking it.
As long as you dont supply any information about your kernel-version and your glibc-version all that remains for us to help you is the crystal ball. Mine is broken at the moment ... by Töns -- There is no safe distance.
BZ writes:
This is the second time I have done this and I just thought the list should know.
Once in awhile (every six months or so) I send Zope a request that it just cannot handle. At that point (after waiting for a good five minutes) I cancel the request by stopping the browser. This has twice resulted in Zope spinning out of control (taking up more and more CPU) untill it kills the entire server box.
How does it kill a Linux box you say?
Well, it kills off Apache, kills off SSH (no more remote access), kills off FTP. They all just become unresponsive. Seems as if Zope eats all network resources...
Here are the two times it has happened:
- Zope 2.3.2, on RH Linux 7.0 384 MB Ram, 650 MHz Pentium: I tried packing the database, stopped it and tried repacking it.
- Zope 2.4.2 (same hardware): I tried creating a VERY LARGE Poll graph (10,000) and stopped it while it got going.
I don't know if this is a bug or whatever, but it could be a problem. Zope can certainly be sent into a tailspin. Usually, you can not stop a Zope request... You can stop your browser waiting for the response, but Zope will usually continue to work on this response to throw it away when it is computed.
But nevertheless, this is not a reason to eat all resources... Dieter
On Sat, 3 Nov 2001 00:04:33 +0100, Dieter Maurer <dieter@handshake.de> wrote:
Well, it kills off Apache, kills off SSH (no more remote access), kills off FTP. They all just become unresponsive. Seems as if Zope eats all network resources...
A recently-fixed ZEO bug can eat file descriptors..... Toby Dickenson tdickenson@geminidataloggers.com
Toby Dickenson writes:
On Sat, 3 Nov 2001 00:04:33 +0100, Dieter Maurer <dieter@handshake.de> wrote:
Well, it kills off Apache, kills off SSH (no more remote access), kills off FTP. They all just become unresponsive. Seems as if Zope eats all network resources...
A recently-fixed ZEO bug can eat file descriptors..... But only with very strange system configuration that of other processes.
Usually, you have a per process limit of 256 or 1024 file descriptors. The operating system should have more, for use by other processes. Dieter
participants (6)
-
BZ -
Chris McDonough -
Dieter Maurer -
Oliver Bleutgen -
Toby Dickenson -
Toens Bueker