Is Yusr working right now?

A queue backlog at Pulsegate, the partner that delivers our push, SMS and email messages, delayed notifications by up to twenty minutes. Payments, transfers and balances were never affected — only the alert that tells you about them. The backlog cleared once the provider drained the queue, and all delayed messages were delivered rather than dropped. We have since split fraud and security alerts onto a dedicated high-priority route so they cannot sit behind marketing traffic.

Authorisation responses slowed from around 90 milliseconds to over four seconds, and roughly 2% of card payments timed out and were declined at the terminal. The cause was connection pool exhaustion in the authorisation service after a scheme certificate rotation left stale connections open. We recycled the pool and payments returned to normal. No customer was charged for a declined payment; a small number of pending holds released within 24 hours as expected. Pool health is now checked after every certificate rotation.

A routing fault in the chat platform sent new conversations to a single agent queue instead of distributing them, so waits stretched past twenty minutes at the busiest point. Conversations already open were unaffected, and the phone line ran normally throughout. Routing was corrected by the vendor and the backlog was worked down by the end of the shift. Customers who left the chat before being answered were called back the same day.

New applications stalled at the document and face-match step after our identity verification partner reported degraded performance on its side. Around 1,900 applications were affected. Existing customers were not impacted at any point. Applicants were able to pick the application back up from where it stopped once the partner recovered, and we messaged everyone who had stalled to let them know they could continue.

Monitoring raised a replication lag alert on the secondary region. Engineers investigated, traced the lag to a long-running index rebuild, and confirmed it cleared on its own once the rebuild finished. No failover was performed and there was no customer impact. Alert thresholds were retuned so a maintenance job of this shape does not page the on-call engineer again.

A bad configuration release caused sign-in to fail on Android with a generic error. iOS was unaffected. We rolled the release back within 39 minutes of the first report. Nobody was locked out permanently and no repeated-failure lockouts were applied to affected accounts.