Merge #149235 #149240

149235: backup: fix flaky OR recovery test r=jeffswenson a=kev-cao

OR recovery test was flaking due to an insufficient timeout and also ran into errors due to the default database being set offline. This patch fixes the test to account for these errors.

Epic: CRDB-51394

Fixes: #148424

Release note: None

149240: process-bep-file: set default server to `mesolite` r=jlinder a=rickystewart

We are migrating workloads from `tanzanite` to `mesolite` as part of merging the two clusters.

Epic: none
Release note: None
Release justification: Non-production code changes

Co-authored-by: Kevin Cao <[email protected]>
Co-authored-by: Ricky Stewart <[email protected]>

Author: 3 people

pkg/cmd/bazci/process-bep-file/main.go

@@ -31,7 +31,7 @@ var (
 	eventStreamFile = flag.String("eventsfile", "", "eventstream file produced by bazel build --build_event_binary_file")
 	jsonOutFile     = flag.String("jsonoutfile", "", "if given, file path where to write the JSON test report")
 
-	serverName    = flag.String("servername", "tanzanite", "URL of the EngFlow cluster")
+	serverName    = flag.String("servername", "mesolite", "URL of the EngFlow cluster")
 	tlsClientCert = flag.String("cert", "", "TLS client certificate for accessing EngFlow, probably a .crt file")
 	tlsClientKey  = flag.String("key", "", "TLS client key for accessing EngFlow")
 

pkg/cmd/roachtest/tests/backup_restore_roundtrip.go

@@ -539,6 +539,12 @@ func testOnlineRestoreRecovery(ctx context.Context, t test.Test, c cluster.Clust
 			)
 		}
 
+		// defaultdb is going to be set offline by the failed download job, so we
+		// need to switch to the system database first to avoid any errors.
+		if _, err := dbConn.ExecContext(ctx, "USE system"); err != nil {
+			return err
+		}
+
 		if _, err := dbConn.ExecContext(
 			ctx, "SET CLUSTER SETTING jobs.debug.pausepoints = ''",
 		); err != nil {
@@ -558,7 +564,7 @@ func testOnlineRestoreRecovery(ctx context.Context, t test.Test, c cluster.Clust
 				err, "waiting for download job %v to reach resumed state", downloadJobID,
 			)
 		}
-		if err := WaitForFailed(ctx, dbConn, jobspb.JobID(downloadJobID), jobStatusWait); err != nil {
+		if err := WaitForFailed(ctx, dbConn, jobspb.JobID(downloadJobID), 10*time.Minute); err != nil {
 			return errors.Wrapf(
 				err, "waiting for download job %v to reach failed state", downloadJobID,
 			)