Sauce Labs Outage History

### **Dates:** Monday, September 28th 2026, 14:19 UTC - 16:52 UTC. ### **What happened:** Real device tests in US West hung and failed to output test artifacts due to slow uploads. ### **Why it happened:** A degraded interconnect link suffered high packet loss from optical errors, but BGP stayed up, preventing automatic failover to healthy paths. ### **How we fixed it:** Manually disabled the degraded port on the edge router, forcing traffic onto healthy alternate interconnects. ### **What we are doing to prevent it from happening again:** Automating link draining based on CRC error thresholds, improving protocols for isolating flapping vendor links, and hardening upload resilience.

**Dates:** Wednesday, September 2nd 2026, 08:30 UTC - 11:15 UTC. ### **What happened:** Real Device Automated Tests experienced failure to start in the EU Central data center, with elevated error rates reaching nearly 100% for short periods of time. ### **Why it happened:** A deployment led to increased cache misses and elevated load on the underlying infrastructure. ### **How we fixed it:** Service was restored by restarting the application. ### **What we are doing to prevent it from happening again:** Additional resources were allocated to reduce the likelihood of a recurrence in the short term. In the long term, we will optimize the performance of resource intensive tasks and introduce dynamic scaling.

### **Dates:** Friday, August 7th 2026, 11:00 UTC - 13:04 UTC. ### **What happened:** Windows and Intel Mac jobs in `us-west1` failed to start due to virtual machine \(VM\) allocation starvation. ### **Why it happened:** A service crash loop left VMs in an allocated but unclaimed state, while a cleanup bug prevented the system from releasing the orphaned capacity to boot new VMs. ### **How we fixed it:** Restored service stability and cleared stale allocations to resume VM provisioning and clear queued jobs. ### **What we are doing to prevent it from happening again:** Fixing allocator cleanup logic, strengthening deployment health checks, and improving capacity accounting for stale allocations.

### **Dates:** Tuesday July 14th 2026, 09:00 - 18:28 UTC ### **What happened:** iOS ARM tests running in our EU data center failed. ### **Why it happened:** A fault occurred with our primary network provider in our EU data center. ### **How we fixed it:** We failed over to a secondary network provider from our EU data center. ### **What we are doing to prevent it from happening again:** We've improved our monitoring and alerting to catch issues with third party network providers.

### **Dates:** Wednesday July 1st 2026, 17:02 UTC - 21:31 UTC ### **What happened:** macOS 14 tests in the US West and EU Central data centers were unable to start. ### **Why it happened:** An internal datasource was unavailable due to a missing configuration entry. ### **How we fixed it:** The missing entry was replaced. ### **What we are doing to prevent it from happening again:** Safeguards have been put in place to prevent in-use datasource removal from configuration.

### **Dates:** Wednesday, June 24th 2026, 09:42 UTC - 15:45 UTC. ### **What happened:** Real Device test sessions using Appium and Access API experienced increased error rates in the US East, US West, and EU data centers. Sessions were timing out after approximately 90 seconds or becoming stuck in the "Connecting" state, preventing them from being closed. ### **Why it happened:** A product defect was introduced, causing Real Device test sessions to fail to start and preventing active sessions from being closed. ### **How we fixed it:** Rollback to a stable version. ### **What we are doing to prevent it from happening again:** Improve monitoring & alerting and enhance post deployment validation.

### **Dates:** Thursday June 18 2026, 03:45 UTC - 06:45 UTC. ### **What happened:** Approximately 7% of iOS devices in our EU data center were temporarily unavailable for customer test sessions after a loss of power to the rack hosting them. ### **Why it happened:** The power circuit feeding the affected rack exceeded its capacity, and a protective breaker tripped to safeguard the line - cutting power to the devices on that rack until the circuit was restored. ### **How we fixed it:** The affected circuit was reset and power restored to the rack, returning the devices to customer service. ### **What we are doing to prevent it from happening again:** We are redistributing power load across affected racks, adding capacity monitoring with early-warning alerts ahead of circuit limits, and introducing a capacity review before new devices are deployed to a rack.

### **Dates:** Monday, June 1st 2026, 13:06 UTC - 15:42 UTC. ### **What happened:** Customers served by the US-EAST-4 region were unable to authenticate or start new test sessions because an incomplete TLS certificate chain was deployed to the core directory services. ### **Why it happened:** A certificate extraction script defect silently truncated the certificate chain after the certificate authority transitioned to a longer hierarchy. ### **How we fixed it:** Reverted the certificate rotation and re-applied the previous known ,good certificate to restore authentication services. ### **What we are doing to prevent it from happening again:** Implementing pre-deployment certificate chain validation, adding active monitoring, and fixing the script's chain-length limitations.

### **Dates:** Wednesday, May 13th 2026, 18:47 UTC - Friday, May 15th 2026, 09:23 UTC. ### **What happened:** Customers were unable to access test run summaries for Real Device Cloud \(RDC\) jobs because events stopped publishing to the jobs Kafka topic in US-EAST. ### **Why it happened:** An authentication key used by the message producer unexpectedly lost its permissions during an account cleanup. ### **How we fixed it:** Manually restored the required permissions to re-establish the connection and resume service. ### **What we are doing to prevent it from happening again:** Migrating to a permanent service account, implementing a Dead Letter Queue \(DLQ\) for the jobs Kafka topic, and replaying the missing events to restore customer data.

### **Dates:** Thursday, April 23rd 2026, 22:43 UTC - Friday, April 24th 2026, 15:29 UTC ### **What happened:** Video assets were missing for virtual iOS simulator tests on ARM and macOS ARM desktop tests in the US-West and EU data centers. ### **Why it happened:** A product defect was introduced resulting in a screen capture failure. ### **How we fixed it:** We performed a rollback to a stable version. ### **What we are doing to prevent it from happening again:** We are improving monitoring & alerting to enhance our post deployment validation.

### **Dates:** Thursday, April 16th 2026, 00:00 UTC – 09:15 UTC ### **What happened:** Live and automated tests on iOS 17.0 simulators failed to start in both the EU and US-West data centers. Customers running tests on iOS 17.0 Intel-based simulators were unable to execute their tests for approximately 9 hours. ### **Why it happened:** A deployment introduced an incompatibility affecting iOS 17.0 on Intel-based infrastructure. The issue was not caught prior to release due to insufficient post-deployment test coverage for that specific simulator configuration. ### **How we fixed it:** We performed a rollback to the previous deployment, which restored full iOS 17.0 simulator functionality. ### **What we are doing to prevent it from happening again:** We are reviving and expanding automated post-deployment tests to cover a broader range of simulator configurations, including legacy Intel-based iOS versions, to catch incompatibilities before they reach production.

### **Dates:** Monday April 7th 2026, ~11:00 – 15:55 UTC ### **What happened:** Some customers experienced 503 errors when running tests via saucectl. The test-composer service was intermittently unavailable, preventing framework-based test execution. ### **Why it happened:** A stale Docker image was deployed to the test-composer service due to a packaging issue that arose during an internal container registry migration. This caused service pods to crash. ### **How we fixed it:** We identified the stale image and redeployed the correct version, restoring the service. ### **What we are doing to prevent it from happening again:** We are hardening our image deployment pipeline and adding validation checks to ensure container registry migrations do not result in stale or incorrect images being deployed to production.

### **Dates:** Tuesday, March 24th 2026, 09:32 UTC – 15:13 UTC ### **What happened:** Network calls failed on iOS devices during Real Device Cloud sessions where network capture was enabled. Approximately 12-13% of iOS sessions were affected. Android was not impacted. ### **Why it happened:** A deployment introduced a DNS resolution change that was incompatible with the iOS platform, causing network capture to break. ### **How we fixed it:** Rolled back the deployment to restore service. ### **What we are doing to prevent it from happening again:** Adding synthetic tests to catch network capture regressions before production, and implementing monitoring alerts for faster detection after deployments.

### **Dates:** Wednesday, March 19 2026, 04:45 UTC - 10:47 UTC. ### **What happened:** Approximately 15% of iOS devices in our US-West data center were temporarily unavailable for customer test sessions due to failed internet connectivity checks. ### **Why it happened:** An automated wireless network optimization feature adjusted transmit power levels on access points serving the affected devices, degrading wireless connectivity and causing devices to fail their availability checks. ### **How we fixed it:** The affected access points were identified and restarted, restoring normal wireless connectivity. ### **What we are doing to prevent it from happening again:** Evaluation of the automated optimization tools and a monitoring improvement.

### **Dates:** Friday, March 13th 2026, 14:43 UTC - 15:11 UTC. ### **What happened:** Real Devices \(iOS and Android\) availability gradually decreased across all data centers. ### **Why it happened:** A product defect was introduced resulting in a small subset of Real Devices \(~10%\) failing to maintain required connectivity. ### **How we fixed it:** Rollback to a stable version. ### **What we are doing to prevent it from happening again:** Improve monitoring & alerting, enhance post deployment validation.

### **Dates:** Tuesday March 10th 2026, 17:52 - 23:34 UTC ### **What happened:** The majority of iOS devices across all regions became unavailable. ### **Why it happened:** Apple's [ppq.apple.com](http://ppq.apple.com) app verification endpoint was down, causing internal device monitoring checks to fail, bringing devices offline. ### **How we fixed it:** We temporarily disabled these device monitoring checks. ### **What we are doing to prevent it from happening again:** Improved external monitoring to catch outages of apple’s [ppq.apple.com](http://ppq.apple.com) endpoint, loosened device monitoring to not take down live iOS devices if [ppq.apple.com](http://ppq.apple.com) is down.

### **Dates:** Friday, March 6th 2026, 21:38 UTC - 23:11 UTC ### **What happened:** During the incident timeline, customers running virtual iOS simulator tests on ARM or macOS ARM desktop tests in the EU Data Center were unable to start new sessions for either live or automated. ### **Why it happened:** There was a sequencing issue on the release of the ARM side disk images in the EU. ### **How we fixed it:** The image reference for the ARM side disk was rolled back to the previous reference to restore service. ### **What we are doing to prevent it from happening again:** The tests that run to validate the image syncing have been completed in each region.