How do I get notified of ECC errors in Linux?
Solution 1:
mcelog will monitor the memory controller and report memory error events to syslog, and in some configurations can offline bad memory pages. This is, of course, in addition to its usual use to monitor machine check exceptions and a variety of other hardware errors.
Most Linux distributions have a service set up to run it as a daemon, e.g. for EL 6:
chkconfig mcelog on
service mcelog start
Solution 2:
The Linux kernel supports the error detection and correction (EDAC) features of some chipsets. On a supported system with ECC the status of your memory controller is accessible via sysfs:
/sys/devices/system/edac/mc
The directory tree under that locations should correspond to your hardware, e.g.:
/sys/devices/system/edac/mc/mc0/csrow2/power
/sys/devices/system/edac/mc/mc0/csrow0/power
/sys/devices/system/edac/mc/mc0/dimm2/power
/sys/devices/system/edac/mc/mc0/dimm0/power
/sys/devices/system/edac/mc/mc1/power
...
Depending on your hardware, you might have to explicitly load the right edac driver, cf.:
find /lib/modules/$(uname -r) -name '*edac*'
The edac-utils package provides a command line frontend and a library for accessing that data, e.g.:
edac-util -rfull
mc0:csrow0:mc#0memory#0:CE:0
mc0:csrow2:mc#0memory#2:CE:0
mc0:noinfo:all:UE:0
mc0:noinfo:all:CE:0
mc1:noinfo:all:UE:0
mc1:noinfo:all:CE:0
You can setup some kind of cron-job that periodically calls eac-util and feeds the results into your monitoring system, where you can then configure some notifications.
In addition to that, running mcelog is generally a good idea. Depends on the system, but uncorrectable/correctable ECC errors are likely reported as machine check exception (MCE), as well. I mean, even brief periods of CPU throttling due to higher temperature are reported as MCE.
Solution 3:
This depends on your server hardware. A whitebox or a Supermicro system will handle this differently than a Dell, HP or IBM...
One of the value-add features of high-end servers is that there's a level of hardware/OS integration. Nicer servers will report what you're looking for as part of the management agents and/or out-of-band management solution (ILO, DRAC, IPMI).
You should use the tools native to your hardware platform.
Excerpt from an HP ProLiant servers running Linux and the HP Management agents:
Trap-ID=6056
ECC Memory Correctable Errors detected.
and
Trap-ID=6052
Advanced ECC Memory Engaged
or a more severe
Trap-ID=6029
A correctable memory log entry indicates a memory module needs to be
replaced.
or the worst... Ignoring an error for 6 days until the server crashes because of bad RAM
0004 Repaired 22:21 12/01/2008 22:21 12/01/2008 0001
LOG: Corrected Memory Error threshold exceeded (Slot 1, Memory Module 1)
0007 Repaired 02:58 12/07/2008 02:58 12/07/2008 0001
LOG: POST Error: 201-Memory Error Single-bit error occured during
memory initialization,
Board 1, DIMM 1. Bank containing DIMM(s) has been disabled.
0008 Repaired 19:31 12/08/2009 19:31 12/08/2009 0001
LOG: ASR Detected by System ROM
These were logged, plus SNMP traps and emails were sent.
Generically, you'll see Machine Check Exceptions in the kernel ring buffer, so you can check dmesg or run mcelog. In my experiences with Supermicro gear without IPMI, that didn't catch everything, and I still had RAM errors slip through the cracks and cause outages. Unfortunately, this led to archaic RAM burn-in policies before system deployments.
Solution 4:
The rasdaemon package was created as a replacement for edac-tools, and newer kernels don't even support edac-tools or mcelog.
An update to the EDAC linux kernel drivers changed how the memory error counters were managed in userspace, so edac-tools and mcelog are effectively deprecated.