Google broke its own cloud again
Software bug confused SSDs into thinking they were full during maintenance
Google has 'fessed up to breaking its own cloud. Again.
The most recent mess occurred on June 28 when Google Compute Engine SSD Persistent Disks in us-central1-a “experienced elevated write latency and errors in one zone for a duration of 211 minutes.” The mess meant that disks probably stopped accepting writes and instances that used SSDs as their root partition probably hung.
Google's very good about revealing just why things go bad in its cloud. This time around it says: “Two concurrent routine maintenance events triggered a rebalancing of data by the distributed storage system underlying Persistent Disk.”
Nothing to worry about there, because “this rebalancing is designed to make maintenance events invisible to the user, by redistributing data evenly around unavailable storage devices and machines.”
Which is just how a cloud should behave: lots of moving parts at the back end invisible to you, who just keeps getting well-behaved servers.
But on this occasion, “a previously unseen software bug, triggered by the two concurrent maintenance events, meant that disk blocks which became unused as a result of the rebalance were not freed up for subsequent reuse, depleting the available SSD space in the zone until writes were rejected.”
And once the disks thought they'd run out of space, no amount of clever back-endery could help for the 211 minutes it took Google to figure it out and set things to rights.
As ever, Google's pledged to do better in future and says its “engineers are refining automated monitoring such that, if this issue were to recur, engineers would be alerted before users saw impact. We are also improving our automation to better coordinate different maintenance operations on the same zone to reduce the time it takes to revert such operations if necessary.”
As we've previously noted, Google is more candid than its rivals when it discloses outages and their causes. But it also appears to have more outages to disclose: The Register monitors the big three clouds' outage notifications and Google announces problems more than either AWS and Microsoft, both of which have larger clouds with more products.
The Alphabet subsidiary's new cloud chief Diane Greene has quite a job ahead of her. ®