Page MenuHomePhabricator

db1101 crashed - memory errors
Closed, ResolvedPublic

Description

db1101, one of the new hosts we recently bought just crashed
I will update this once I get more information.

The server was executing some heavy ALTER tables.

From the logs:

/admin1/system1/logs1/log1-> show record1

	properties
		CreationTimestamp = 20171017114511.000000-300
		ElementName = System Event Log Entry
		RecordData = Multi-bit memory errors detected on a memory device at location(s) DIMM_A4.
		RecordFormat = string Description
		RecordID = 5

/admin1/system1/logs1/log1-> show record3

	properties
		CreationTimestamp = 20171012141437.000000-300
		ElementName = System Event Log Entry
		RecordData = Correctable memory error rate exceeded for DIMM_A4.
		RecordFormat = string Description
		RecordID = 3

Related Objects

Event Timeline

Marostegui moved this task from Triage to In progress on the DBA board.
Marostegui changed the visibility from "Public (No Login Required)" to "WMF-NDA (Project)".Oct 17 2017, 12:53 PM
Marostegui updated the task description. (Show Details)
Marostegui renamed this task from db1101 crashed to db1101 crashed - memory errors.Oct 17 2017, 12:55 PM
Marostegui updated the task description. (Show Details)
Marostegui changed the visibility from "WMF-NDA (Project)" to "Public (No Login Required)".
Marostegui added a subscriber: • Cmjohnson.

@Cmjohnson can we get a new dimm for this host to replaced that one mentioned on the logs?
Thanks

I have powercycled the host and it came back up fine, but we better replace that DIMM as the server is quite new.
Going to execute the alters again to see if it crashes once more.

The ALTER tables finished correctly and no more crashes happened.
Let's change the memory anyways @Cmjohnson

A new DIMM has been requested with Dell.

You have successfully submitted request SR955416674.

Mentioned in SAL (#wikimedia-operations) [2017-10-25T15:31:12Z] <marostegui> Power off db1101 for HW maintenance - T178383

Dell declined to send the new DIMM, stated that my supporting documentation was insufficient. I swapped the DIMM at A4 to B4 and will need to wait for that to fail before submitting again.

How's that possible? Aren't the logs from _their_ server's idrac enough??

Marostegui closed this task as Resolved.EditedOct 25 2017, 5:03 PM

Anyways as per my chat with Chris, about Dell's procedures - I will mark this as resolved (mysql is back up) and let's reopen once it fails again.
I will do some heavy alters in the next few days so we will see...

Thanks Chris!

I have started 16 (1 per database present ) concurrent alters for the templatelinks table to generate some load

And now almost 50 alters running at the same time

Mentioned in SAL (#wikimedia-operations) [2017-10-26T14:10:15Z] <marostegui@tin> Synchronized wmf-config/db-eqiad.php: Repool db1101 - T178383 (duration: 00m 49s)