Red Hat Bugzilla – Bug 909186
RPM has an excessive HDD usage
Last modified: 2015-11-23 22:51:26 EST
Description of problem: When I do 'yum update' yum rebuilds RPM 'Packages' database many times over, writing out up to 800MB (!) of data to the disk (whereas my 'Packages' file weighs just 35MB). That's insane. Version-Release number of selected component (if applicable): yum-3.2.29-30.el6.noarch rpm-4.8.0-27.el6.i686 How reproducible: Always, specially if I install/update Google Chrome Additional info: I've tried to rebuild the RPM database several times, but it had no effect whatsoever. I can upload my entire /var/lib/rpm if anyone's interested (a 10MB tar.xz archive).
OK, this is getting ridiculous. I've just run yum search dhcp and according to iotop, this operation alone has made yum write out over 400MB of data to the disk. Please, fix this insanity.
As a temporary workaround I've resorted to 1) cp -a /var/lib/rpm /tmp (tmp is tmpfs) 2) do everything I need to with yum 3) cp -a /tmp/rpm /var/lib
Can you at least save /var/lib/rpm somewhere it won't go away. If you can get an strace/ltrace that might be helpful. Also if you manually do: "rpm --rebuilddb" ... does that solve the problem?
(In reply to comment #4) I mentioned in the original message that the rebuilding of the RPM database hadn't helped. strace shows billions of write() Anyways, here's my entire RPM db for your consideration: ompldr.org/vaGZzbQ (sha1sum: c458f9a089e0c2f946307eec3c337a0381e4e51e) I've changed the bug report title, since `rpm -Uvh package.rpm` causes the same problem.
You are most likely not measuring correctly. Even if you would measure correctly the only way around this behaviour would be replacing the rpmdb. Work like this has just started upstream but is far from being finished. Even if upstream ends up replacing the current rpmdb implementation backporting such massive changed to RHEL6 is completely out of question. Closing.