-
Type:
Bug
-
Resolution: Unresolved
-
Priority:
Major - P3
-
None
-
Affects Version/s: None
-
Component/s: None
-
Server Security, Server Triage
-
ALL
-
None
-
None
-
None
-
None
-
None
-
None
-
None
As detailed in https://jira.mongodb.org/browse/HELP-88974, for MONGODB-AWS authentication, the aws_sts_request auth metric was reported as micros=0.
From the slow query logs, we can see that the actual cpu time for the auth handshake was "cpuNanos":9963328 (10ms) but the step 2 (server processing the saslContinue from the client) time was "duration_micros":15725222 (15.7s), which suggested the saslContinue spent time waiting on the server to STS GetCallerIdentity call.
Analyzing the code for the mongod log her: https://github.com/10gen/mongo/blob/master/src/mongo/db/modules/enterprise/src/sasl/sasl_aws_server_conversation.cpp#L45, it appears that the timer used to populate aws_sts_request in the log is not wrapping the http request so it doesn't capture the true latency of the STS call.