Between 21:30 UTC 25 November 2019 and 07:05 UTC 26 November 2019 the Websocket API was running with degraded capacity. This resulted in slightly more latent feeds during this time and some isolated occurrences of substantial lag during traffic spikes correlated with large market moves.
The issue was identified when latency thresholds were breached in our automated monitoring systems around 06:55 UTC. From 07:05 the issue was resolved and full capacity was restored. We continue to monitor the impacted services closely.
The issue was caused by incorrect CPU pinning following a release of a market data distribution component at 21:30 UTC 25 November 2019. The impact of this was only observed during large traffic spikes which occurred several hours later and so was not identified during the post release checks at the time. The configuration of this service has been corrected and the deployment tested to prevent this from happening again.