mirror of https://github.com/OpenVidu/openvidu.git
4d6b7acd7c
Parallel master creation: drop the MasterNodeN -> WaitConditionN-1 DependsOn chain. Each master gets a fixed MASTER_NODE_NUM via UserData (replaces subnet-matching detection) and publishes its private IP to its own SSM parameter (atomic per-key writes; the 4 parameters are stack resources), removing the IP exchange through the shared Secrets Manager JSON and its lost-update race. after_install.sh is now leader-only, so the shared JSON keeps a single writer with parallel masters. Ungate media nodes and the NLB from WaitCondition4: media wait for secrets + the 4 SSM IPs (bounded polls) before the heavy install, and for a healthy master (:7880/health/caddy) before starting the service. Bounded timeout on the previously infinite IP wait, robust installer fetch (curl --retry to file instead of sh <(curl)), SSM read/write IAM statements, MEET_INITIAL_API_KEY dead-code fix, SubnetProcessor Lambda cleanup (unused boto3 client and ec2:DescribeSubnets permission). Validated with ov-cloud-tester (sc-deploy-destroy, ha, dev): PASS. deploy 5m24s (baseline ~12m), ready 21m14s (wait-ready dominated by external DNS propagation), destroy 7m39s. |
||
|---|---|---|
| .. | ||
| DOC-CHANGES-ha-optimizations.md | ||
| cf-openvidu-ha.yaml | ||