bryancall opened a new pull request, #13872: URL: https://github.com/apache/trafficserver/pull/13872
`stek_share` started five ATS nodes, slept 10 seconds, and then expected every node to resume one TLS session. Each node checks for the leader's session ticket key when it joins the Raft cluster and then only every 10 seconds. A node whose first check comes before the key arrives installs it about 10 seconds later, and a leader elected after its own first check sends its key only on the next one. The handshakes started about 10.2 seconds after the last node did, so on a busy CI agent they could beat a node's key install and the resumption check failed. This replaces the fixed sleep with a wait on each node's log line for the shared key, so the handshakes start only once every node is using it. Testing, in the fedora:44 CI image under CPU load: 960 runs of master failed 2 times with the CI signature and 960 runs of this change failed none. Cutting master's sleep from 10 to 9 seconds makes it fail 60 of 80 runs, which shows how close the old wait sat to the key refresh. Breaking key sharing by moving one node's Raft port makes the new test fail 8 of 8, so it still catches a cluster that does not share its key. -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected]
