Half your services start timing out at once and nothing was deployed. How do you run that incident?

Declare the incident, assign an incident commander, and separate investigation from communication. Then hunt the shared dependency: authentication, DNS, a database, a certificate, the service mesh control plane, or a cloud zone. Check what changed outside deploys, meaning config pushes, feature flags, DNS records, expiring certs, and provider status. Mitigate before you diagnose by shedding load, failing over, or disabling the suspect flag, then verify with user facing SLIs.

Broad and simultaneous with no deploy means it is almost never ten independent bugs, it is one thing underneath all of them. So I get incident structure up immediately, commander and a comms person, then start on the shared layer list: auth service, DNS, the primary database, the mesh control plane, certificates, and cloud provider health. In parallel I ask what changed that is not code, because config pushes and feature flags are changes people forget to count. We had exactly this pattern once and it turned out to be an internal certificate authority renewal that had quietly expired an intermediate, so every mutual TLS handshake started failing at the same minute. The tell was that failures were connection level rather than application level. While that investigation runs I want a mitigation lever ready, usually shedding non essential traffic or failing over to the secondary region, because I would rather be degraded and stable than fully broken while we think. Then verify with real user metrics, not just green pods.