Cases10 cases

6 to 10 of 10

1. IL-0308 Most RevenueCat backend systems failed after cutover to a new Aurora PostgreSQL 14 cluster as queries became very slow

   RevenueCat2022

   Symptoms

   At 9:34 UTC on November 23, 2022, RevenueCat started moving traffic from its Aurora PostgreSQL 10.x cluster to a new Aurora PostgreSQL 14.x cluster. By 9:39 UTC all traffic was on the new cluster, latency was bad, and the RevenueCat dashboard and the endpoints used by the RevenueCat SDK stopped working.

   Database overloadHigh latencyService outageConnection pool exhaustion

   Trigger

   At 9:34 UTC production traffic was switched from the Aurora PostgreSQL 10.x cluster to the new Aurora PostgreSQL 14.x cluster, before ANALYZE had run on its largest tables.

   Root cause

   Breaking change in an upgrade

   Component

   DatabasePostgreSQL vacuumDatabase replicationDatabase proxy and connection poolerQuery planner and table statistics

   Contributing factor

   Unrepresentative stagingUnverified assumptionAutomation amplificationMonitoring blind spot

   Detection

   At 9:39 UTC, with all traffic on the new cluster, latency looked bad and the infrastructure team began to investigate.

   Remediation

   • Added a statement timeout to the database user so queries could not hold a connection too long
   • Shed 90% of traffic by returning 503s from the load balancer
   • Dropped indexes that the planner was using wrongly and that were not needed
   • Stopped accepting Apple webhooks for a time
   • Rolled out a release without New Relic to stop its EXPLAIN queries
   • Ran ANALYZE by hand on the biggest tables
   • Raised the values of the affected primary key sequences with a prepared script
   • Stopped the integration event dispatcher by hand
   • Replayed the webhooks that were dispatched but not recorded
   • Restored integration event dispatching, ETLs and other non-web server jobs

   Action items

   • Consider change freezes during business-critical periods
   • Explain better how the SDK behaves during backend outages
   • Tell customers about maintenance and outages in the developer dashboard and by direct email to technical contacts
   • Replay a subset of write traffic in migration dry runs to mimic the cutover
   • Move costly processes to asynchronous work to keep the API fast and stable
   • Re-evaluate table indexes, including unused ones the database may choose for queries
   • Show the load of each endpoint and system in the database dashboards
   • Involve support and customer communications staff during outages

   Lessons

   The new cluster had not been analyzed, and the tests missed it because the test cluster had been up long enough to gather statistics. Dry runs replayed no write traffic, on the belief that writes are simpler than reads, and the team assumed sequence state was replicated with the data. Rolling back at the start would have been better than fixing forward. The migration, timed to finish before Thanksgiving and Black Friday, should have been delayed until the week after, because the US had entered a sensitive period and part of the US support team was already on holiday.

   View case
2. IL-0311 BattleMetrics' primary PostgreSQL went read-only at transaction ID wraparound, breaking the website and RCON updates

   BattleMetrics2022

   Symptoms

   At 03:23 UTC on 27 March 2022, BattleMetrics' primary PostgreSQL database, which holds all account, server and player information, went into read-only mode. Users first saw the website fail to load, and soon RCON updates and queries began to fail as well.

   Writes rejectedService outage

   Trigger

   A corruption of the index on the player identifiers table, which BattleMetrics' logs place at about 26 February 2022, made every vacuum job fail, so transaction IDs kept climbing until Postgres stopped accepting writes on 27 March.

   Root cause

   Data loss or corruption

   Component

   DatabasePostgreSQL vacuumDatabase replicationDatabase backups and snapshots

   Contributing factor

   Monitoring blind spotOverlapping failuresNo headroom

   Detection

   Users failed to load the website and reported it as their API keys expired, then RCON updates and queries began to fail, and the team found that the main database no longer accepted writes.

   Remediation

   • Shut the database down and dropped the corrupted player identifiers index in single-user mode
   • Ran the vacuum again, which succeeded and let the database accept writes
   • Rebuilt the player identifier index, which took about two hours
   • Shut down the query workers and gave RCON updates priority
   • Pointed services at the new Postgres primary
   • Bootstrapped new replicas one after another to take load off the primary
   • Restarted queries with throttling, prioritised by server population and rank, after RCON updates had caught up

   Action items

   • Monitor errors in the Postgres logs
   • Monitor and alert on transaction IDs
   • Set up a logical replica in addition to physical replication
   • Build a faster network to speed up recovery from failure

   Lessons

   BattleMetrics names skipping health checks on transaction ID size as its critical mistake: the index corruption silently broke vacuum for a month, and without monitoring of transaction IDs nobody saw the database approach wraparound. Because the corruption had been replicated to the replicas, failing over did not help, and the backup servers and downloadable backups carried the same corruption. The team also found that, contrary to the error message, a full vacuum on a server in wraparound protection does not need single-user mode, and chose to run a slow, delayed service over no service while replicas were rebuilt.

   View case
3. IL-0443 A Repro Aurora MySQL reader restarted unexpectedly and writer insert latency rose about sevenfold for a day or more

   Repro2019

   Symptoms

   A reader of Repro's Aurora MySQL cluster restarted for no known reason, and right after that the writer's insert latency rose to about seven times normal and writer performance stayed badly degraded for a long time. This happened more than once, twice within two weeks by 11 November 2019. While the writer was slow, the applications that process data collected by the Repro SDK fell behind, and the team could only watch until the writer recovered.

   High latencyReplication lagThroughput degradation

   Trigger

   Long-running queries on the reader; the author recalls that the cause was a query sent to the reader from Redash.

   Root cause

   Capacity exhaustion

   Component

   DatabaseDatabase replication

   Contributing factor

   Monitoring blind spot

   Remediation

   • Waited for writer performance to return, since the rollback segment release could not be sped up

   Action items

   • Add a monitor on the rollback segment, mysql.innodb.history_list_length in Datadog
   • Kill queries that run long enough to make the reader restart as soon as they are found
   • Set max_execution_time so queries are stopped after a fixed time

   Lessons

   Once the rollback segment has grown large, nothing speeds up its release across the cluster, so the way to avoid the latency is to keep transactions small and stop long-running queries before the rollback segment grows.

   View case
4. IL-0334 Instapaper's RDS MySQL bookmarks table hit the 2TB ext3 file size limit, inserts failed and the service went down

   Instapaper2017

   Symptoms

   On Wednesday, February 9, 2017 at 12:30PM PT, inserts of new entries into the bookmarks table of Instapaper's MySQL database on Amazon RDS started to fail with an error. The bookmarks table, which stores the articles Instapaper users save, had grown past 2TB, and the Instapaper service was down until Thursday, February 10 at 7:30pm PT.

   Writes rejectedService outage

   Trigger

   At 12:30PM PT on February 9 the bookmarks table grew past the 2TB file size limit of the instance's filesystem.

   Root cause

   Capacity exhaustion

   Component

   DatabaseDatabase backups and snapshotsDatabase replication

   Contributing factor

   Unknown limitMonitoring blind spotUntested backupNo runbookLimited debug access

   Remediation

   • Dumped the production database to a 5.5TB RAID-0 disk on an i2.8xlarge instance with Pinterest's SRE team
   • Launched a temporary production instance with limited access to archives
   • Set up an Aurora read replica of the old database as one recovery path
   • Imported the data into a new MySQL RDS instance without secondary indexes, then built the indexes
   • Had an Amazon engineer rsync the data from the ext3 filesystem to an ext4 filesystem mounted on the failed database
   • Synced the new ext4 database with the temporary production database through row-based replication
   • Promoted the new database to master and pointed the application at it

   Action items

   • Define a workflow that escalates system-wide Instapaper outages immediately to Pinterest's SRE team
   • Test MySQL backups every month instead of every three months
   • Put a process in place with Pinterest's SRE team to recover from this type of failure

   Lessons

   A filesystem limit that Instapaper did not know about and could not see made both the production database and all its backups useless, and the only way out was to restore the data onto a new filesystem. Because the hosted database could only be reached through MySQL, filesystem-level fixes such as rsync needed Amazon engineers. Even with a perfect response, rebuilding the database would have taken at least 10 hours. The lack of a disaster recovery plan for this kind of failure, and the estimate of the dump time from row counts, made the downtime and recovery longer than necessary, and the response could have drawn better on Pinterest's and AWS's people. The new escalation workflow and monthly backup tests would not have prevented this failure but would speed up the response to a future one.

   View case
5. IL-0296 Transloadit's us-east Redis cluster lost queued jobs and heartbeats, raising WORKER_JOB_ERROR errors for 81 customers

   Transloadit2016

   Symptoms

   Reports came in of more WORKER_JOB_ERROR errors, which tell Transloadit there is a communication problem between its Uploaders, the machines that receive files and orchestrate Assemblies, and its Drones, the machines that run encoding tasks. Many Assemblies became much slower because retries with exponential back-off kicked in, and a few could not be replayed.

   High error rateHigh latencyLoad imbalance

   Trigger

   A steep increase in HLS traffic, together with more Drones online processing the same volume faster, which put more load on the central Redis nodes.

   Root cause

   Capacity exhaustion

   Component

   In-memory cacheDatabase replication

   Contributing factor

   Monitoring blind spotUnverified assumptionHarmful manual action

   Detection

   Reports of an increased number of WORKER_JOB_ERROR errors.

   Remediation

   • Re-added lost messages and relaxed heartbeat monitoring as a stopgap
   • Upgraded Redis to the latest stable version
   • Doubled the Redis cluster to six nodes and then added three more, which made it worse
   • Brought the number of read replicas down to one, which cut traffic and failure rates
   • Capped how far Drones scale up in each region

   Action items

   • Add logging and alarms for approaching saturation
   • Find ways to decentralise the traffic between the Redis nodes and the Drones and Uploaders
   • Add sharding per Uploader on top of the sharding per region

   Lessons

   To handle more traffic Transloadit had to use fewer Redis nodes, not more: each added replica multiplied the replication traffic from the single leader that takes every write, so scaling up the number of nodes was the wrong call. Bandwidth had not been among the monitored Redis metrics.

   View case

By root cause6