m/test/launch/cluster: fix duplicate NodeIDs entries
This code loop looks for any NEW nodes and inserts them into
cluster.{Nodes,NodeIDs}. The first structure is a lookup from NodeID to
node information, the latter is a list of NodeIDs used to look up
numeric node IDs (in order of startup/detection) to NodeIDs.
The loop in the code runs multiple times to catch any NEW nodes might
take a while to appear, and exits once the expected number of nodes have
been detected.
The bug caused the code to repeatedly insert into Nodes[NodeID] (which
is fine, but wasteful) and into NodeIDs. The latter resulted in a
NodeIDs that contained duplicate entries.
To make sure we only handle each NEW node once, we skip nodes that have
already been seen.
This bug caused us to sometimes act on wrong nodes in E2E tests (any
time tests were looking up node number -> node ID -> Node via .NodeIDs
and .Nodes, they had a chance of picking the wrong node ID / node for a
given node number).
Change-Id: Ie459c8277c0d03902ce23f3b20b0c4e367cc015b
Reviewed-on: https://review.monogon.dev/c/monogon/+/2881
Tested-by: Jenkins CI
Reviewed-by: Lorenz Brun <lorenz@monogon.tech>
diff --git a/metropolis/test/launch/cluster/cluster.go b/metropolis/test/launch/cluster/cluster.go
index e73385d..14a7307 100644
--- a/metropolis/test/launch/cluster/cluster.go
+++ b/metropolis/test/launch/cluster/cluster.go
@@ -973,6 +973,9 @@
if n.State != cpb.NodeState_NODE_STATE_NEW {
continue
}
+ if seenNodes[n.Id] {
+ continue
+ }
seenNodes[n.Id] = true
cluster.Nodes[n.Id] = &NodeInCluster{
ID: n.Id,