On Mon, Dec 20, 2010 at 12:44, Ken Schumacher <[email protected]> wrote: > Stephen, > > > On Dec 17, 2010, at 7:02 PM, Stephen John Smoogen wrote: > >> On Thu, Dec 16, 2010 at 14:28, Ken Schumacher <[email protected]> wrote: >>> Greetings, >>> >>> I have a repeatable problem on at least one of our SLF 4.4 systems. It >>> seems that running commands like 'yum --check-update' seem to run into some >>> sort of memory leak. The yum output gets to the point of saying "Reading >>> repository metadata in from local files" and a top listing on a another >>> window shows the memory use simply climbing. The original window will not >>> respond to a Ctrl-C. >> >> 1) various versions of Yum does not respond to Ctrl-C because doing so >> can cause the rpm package database to be left in a bad place. > > That's inconvenient in my current situation, but I understand the thinking > behind it. I can work around this by having a second window open allowing me > to 'kill -15' the yum process once it gets into this bad state. > >> 2) Yum will use a lot of memory depending on how much is installed. Of >> course a lot is subjective and needs to be quantified. [100 mb was a >> lot on one system and nothing on another.] > > I wait about 60 CPU seconds before killing the yum process. According to > 'top', at that point it is using 100% of one CPU and it has already allocated > itself 2 GB of memory. On this cluster head node, that is just a bit over > 10% of the node's memory, but I am concerned about letting it go on consuming > memory for fear of interfering with other services on the node. > > I have checked the version of the yum and yum.conf RPMs on this node and > compared to other systems we maintain. We have other systems running those > same versions without this memory consumption problem. I have run yum using > the '-d 5' flag to get some verbose debug output. The last output before > this memory consumption starts says:
Would need to know what is installed on the system > Reading repository metadata in from local files > Setting up Package Sacks > >> 3) 4.4 is really old. 4.8 is standard now and 4.9 will be out of the >> door by summer (it will also probably be the last 4.x series like the >> 3.9 was the last of the 3 series.) > > The node was originally installed with the LTS 4.4 release (Wilson). Until > recently, we have been running daily yum updates against the node, so all the > necessary errata and security updates have been applied. Being a cluster > head node, we can't jump the node up to a 5.x release without proper planning > and scheduling of downtime, etc. Our user base expects the release to remain > stable, so such upgrades are carefully considered. Well I thought that applying all the updates would bring the system to 4.8 but I realize that Scientific Linux does keep old releases alive. What does 'rpm -Va --nofiles' tell you? How do you get the repodata for the systems (local mirror or remote one)? Can you try updating 1-2 packages directly? or does even yum list give you a 2GB process? -- Stephen J Smoogen. "The core skill of innovators is error recovery, not failure avoidance." Randy Nelson, President of Pixar University. "Let us be kind, one to another, for most of us are fighting a hard battle." -- Ian MacLaren
