All times are shown in UTC
At 17:00 a necessary and unplanned migration was performed on our primary website database causing the website to be unavailable for 1-2 minutes.
Our health check system has reported this issue as resolved.
We will continue to investigate the iss0ue and will update this incident shortly.
Resolved
41 minutesThis incident was created automatically by our automated health check system as it has identified a fault. We are now looking into this issue.
We have identified an issue where some HTTP requests are intermittently stalled in the load balancer in the us-west-1 region.
The issue only affects us-west-1. As a precaution we are temporarily routing traffic away from that region. During this period clients that would have used that region will mostly now end up in us-east-1 (East Virginia). Aside from a possibly slight latency increase this should not cause any problems.
6th Jun 11:39 PMWe were hoping that AWS would be able to resolve the issue in a sensible time so that they could diagnose the underlying problem on their end. However, as this did not happen time, we moved forwards and recycled the necessary load balancers and brought the US West region back online. All services are back to normal.
Resolved
3 minutesToday, 1 June 2018, at 13:00 UTC we will be performing essential maintenance on our Sydney datacenter. During this time, all traffic will be routed to the nearest other Ably datacenters , which will most likely be Singapore and California.
We expect this upgrade to take 15-30 minutes, and once completed, all traffic will be routed back to Sydney.
The maintenance is unfortunately taking longer than predicted as we are waiting on AWS to re-provision resources in that region. Once their re-provisioning is complete, the region will be back online.
1st Jun 01:48 PMThe Sydney datacenter is now back online and operating normally.
Resolved
about 1 hourWhilst upgrading a number of RabbitMQ nodes, we had a hardware failure on one node. Whilst all nods are configured in a HA configuration, we had some issues introducing new nodes into the system which caused availability issues of some queues for short periods of up to 30 seconds. This happened a few times between 12:00 UTC and 1:20UTC.
The issues are now fully resolved.
Resolved
less than a minuteSome of the nodes serving the website failed at 7:13 today.
Website issues were resolved by an automated restart of the Heroku dyno. The core realtime system does not use Heroku and was unaffected.
Resolved
29 minutesWe have seen a significant and sudden increase in us-east-1 since 23:30 UTC today.
We are manually over provisioning the capacity in this region and will be investigating the cause of this sudden increase in load. Once identified, we'll look into how we can more effectively prepare for spikes like this.
Unfortunately after provisioning more capacity, a second wave of traffic has arrived.
We are provisioning again.
We have identified the customer who is largely responsible for this traffic spike and have resolved the issue in the affected regions. Over the next few days we'll be adding additional limits to prevent usage patterns like this impacting other accounts in the multi-tenanted cluster.
Resolved
8 minutesBetween 15:57 UTC and 16:02 UTC the EU West 1 (Ireland) cluster experienced a sudden increase in load which caused error rates to climb temporarily in this region.
We believe all HTTP and connection requests should have been retried automatically in other regions using our fallback capabilities.
Resolved
less than a minuteWe are seeing intermittent higher latencies in both US regions at present which is affecting a small number of publishes. The intermittent latency delays we are seeing are in the range of 50-500ms.
We are investigating the root cause of the issue and are actively working to resolve this.
In this case the latencies returned to normal after a redeploy of affected instances. The underlying cause is believed to be a recent regression that leaks object references and causes certain data structures to grow over time. A fix is being prepared and will be deployed in due course.
Resolved
1 minuteSince 10:40 we are seeing elevated error rates in the eu-central-1 (Frankfurt) data centre, and are investigating.
All other regions are operating normally.
Since 10:50, other regions (in particular us-east and eu-west) are experiencing increased latencies as a result of problems in eu-central.
We have now shut down the eu-central-1 region and are redirecting traffic to eu-west-1
5th Apr 11:51 AMError rates have returned to normal and we are continuing to investigate the root cause of the issue that arose in Frankfurt (eu-central-1)
5th Apr 12:17 PMThe cluster has been stable so we consider this issue resolved. We will now continue to investigate the root cause and conduct a post-mortem of the issue.
Resolved
less than a minuteHeroku's API is unavailable due to unexpected maintenance which has resulted in our website being unavailable as this is hosted with Heroku.
We are looking into a solution, which is challenging given the API is down.
See https://status.heroku.com/incidents/1459
Please note that whilst our website is unavailable, this has absolutely no effect on our realtime platform which continues to run without fault.
26th Mar 11:43 PMThe Heroku API has come back online and all services are operating normally again.
Resolved
less than a minuteFor the last few days, a bug related to protocol header forwarding resulted in connections to the Pusher translator over SSL (that is, where `encrypted: true` was set in the Pusher client lib constructor) to be incorrectly rejected with the following error: "Invalid use of Basic authentication over non-TLS transport (Ably error 40103)".
We apologise for the length of time it took for this to be fixed and will be instituting steps to make sure it doesn't happen again. We invite affected customers to get in touch with us at support@ably.io .
Resolved
38 minutesWe are investigating increased latencies in ap-southeast-2 since 15:39 UTC
Update: Latencies and error rates are back to normal as of 15:53. Regions outside of ap-southeast-2 should have been relatively unaffected except when communicating with clients connected to ap-southeast-2. The cause was a new version of our backend, which showed no issues in CI or staging environments. We will update this incident once we discover the root cause.
Resolved
3 minutesWe have seen latencies rise twice today for a period of roughly 2 minutes.
We have identified the root cause of the problem and will aim to roll out a fix to reduce the affect of noisy neighbours on all customers.
Resolved
13 minutesWe experienced increased error rates across the production cluster starting at 1015UTC today, due to failing instances in us-west. The impact was mainly on the us-west region but certain other apps and accounts were affected globally.
New instance were brought online qt 10:28 and performance/error rates were back to normal at 10:30.
Following a review of the impact, we can see that error rates elevation was most prevalent in us-west-1 (California), however we unfortunately also experienced some intermittent increased error rates in us-east-1 (North Virginia) and ap-southeast-2 (Sydney) during this time.
Resolved
about 9 hoursHealth issues with one of our Cassandra nodes in US-East (our persistance and storage layer) is leading to performance issues (and some "Cannot achieve consistency level QUORUM" errors with account/app creation or alteration actions, or message publishes to channels on which persistance is enabled), especially in US east and US west
One of the Cassandra nodes in US-east appears to have become partitioned from other nodes, though the realtime service was still connected to it. This resulted in any queries reaching that node failing with a "Cannot achieve consistency level QUORUM" or "Cannot achieve consistency level TWO" error. This was resolved at 16:53 by shutting down Cassandra on the affected node, since there is sufficient redundancy for the cluster to run fine with one fewer node.
All services are fully back to normal; we're now investigating the root cause of the netsplit.
Resolved
less than a minute