Incident documentation/2021-10-08 network provider

From Wikitech-static
Jump to navigation Jump to search

document status: in-review

Summary

Around 16:11 UTC our (non-paging) monitoring and users reported connectivity issues to and from our Eqiad location. Traceroutes showed a routing loop in a provider's network.

At 16:19 UTC, using the provider's APIs, we asked the provider to stop advertising the prefixes for Eqiad on our behalf.

At 16:24 UTC, the first reports of recoveries arrived as the change propagated through the DFZ.

Unfortunately, due to the preponderance of the Eqiad impact and its recovery, we didn't notice that it also impacted users from Russia to reach the wikis through Esams. We also didn't receive any NEL reports from most Russia users until 16:24 or later, as the location we use for NELs about Esams is itself Eqiad.

At around 17:15 UTC, our monitoring shows a full recovery, indicating the issue being resolved upstream.

Impact: For upto an hour, some regions experienced a partial connectivity outage. This primarily affected the US East Coast for ~13 minutes, and Russia for 1 hour. A subset of readers and contributors from these regions were unable to reach any wiki projects. Services such as Phabricator and Gerrit Code Review were affected as well. It was a partial issue because the network malfunction was limited to one of many providers we use in the affected regions.

Due to the span of this provider's network, the further a client is from Eqiad the more likely they will be using that provider to reach our network and thus could be impacted.

Documentation:

Actionables