aboutsummaryrefslogtreecommitdiffstats
path: root/controller-server/src/main/java/com/yahoo
diff options
context:
space:
mode:
authorJon Marius Venstad <venstad@gmail.com>2022-01-19 22:44:36 +0100
committerJon Marius Venstad <venstad@gmail.com>2022-01-19 22:44:36 +0100
commit0db0da43c891e1869247f6c2738bbbd4fcee2835 (patch)
tree4f988ef7d88ba51fa7e35b63d02128bf1dad3272 /controller-server/src/main/java/com/yahoo
parentc9fb494b594c6f3f2d6d2000a6eb649571fba54c (diff)
Let deployment run for some time before giving up nodes which are down
Diffstat (limited to 'controller-server/src/main/java/com/yahoo')
-rw-r--r--controller-server/src/main/java/com/yahoo/vespa/hosted/controller/deployment/InternalStepRunner.java14
-rw-r--r--controller-server/src/main/java/com/yahoo/vespa/hosted/controller/deployment/Run.java2
2 files changed, 7 insertions, 9 deletions
diff --git a/controller-server/src/main/java/com/yahoo/vespa/hosted/controller/deployment/InternalStepRunner.java b/controller-server/src/main/java/com/yahoo/vespa/hosted/controller/deployment/InternalStepRunner.java
index 225634634b2..a6d1800bf71 100644
--- a/controller-server/src/main/java/com/yahoo/vespa/hosted/controller/deployment/InternalStepRunner.java
+++ b/controller-server/src/main/java/com/yahoo/vespa/hosted/controller/deployment/InternalStepRunner.java
@@ -349,14 +349,12 @@ public class InternalStepRunner implements StepRunner {
String failureReason = null;
- NodeList suspendedTooLong = nodeList
- .isStateful()
- .suspendedSince(controller.clock().instant().minus(timeouts.statefulNodesDown()))
- .and(nodeList
- .not().isStateful()
- .suspendedSince(controller.clock().instant().minus(timeouts.statelessNodesDown()))
- );
- if ( ! suspendedTooLong.isEmpty()) {
+ NodeList suspendedTooLong = nodeList.isStateful()
+ .suspendedSince(controller.clock().instant().minus(timeouts.statefulNodesDown()))
+ .and(nodeList.not().isStateful()
+ .suspendedSince(controller.clock().instant().minus(timeouts.statelessNodesDown()))
+ );
+ if ( ! suspendedTooLong.isEmpty() && deployment.get().at().plus(timeouts.statelessNodesDown()).isBefore(controller.clock().instant())) {
failureReason = "Some nodes have been suspended for more than the allowed threshold:\n" +
suspendedTooLong.asList().stream().map(node -> node.node().hostname().value()).collect(joining("\n"));
}
diff --git a/controller-server/src/main/java/com/yahoo/vespa/hosted/controller/deployment/Run.java b/controller-server/src/main/java/com/yahoo/vespa/hosted/controller/deployment/Run.java
index 2b9e3dd0733..4b38306f905 100644
--- a/controller-server/src/main/java/com/yahoo/vespa/hosted/controller/deployment/Run.java
+++ b/controller-server/src/main/java/com/yahoo/vespa/hosted/controller/deployment/Run.java
@@ -210,7 +210,7 @@ public class Run {
return lastVespaLogTimestamp;
}
- /** Returns the timestamp of the last time no nodes were allowed to be down. */
+ /** Returns since when no nodes have been allowed to be down. */
public Optional<Instant> noNodesDownSince() {
return noNodesDownSince;
}