From: Brad Hubbard <bhubbard@redhat.com>
Date: Tue, 16 Jul 2024 02:09:01 +0000 (+1000)
Subject: qa: Restrict rados api tests to large clusters and increase timeout
X-Git-Tag: v20.0.0~1362^2
X-Git-Url: http://git-server-git.apps.pok.os.sepia.ceph.com/?a=commitdiff_plain;h=d034fec463330d9cefba05bfe9beb0c066e03b43;p=ceph.git

qa: Restrict rados api tests to large clusters and increase timeout

Running these tests with thrashers on small clusters leads to many very
slow ops due to the cluster being overloaded. That has a tendency to
make some of the API tests timeout and fail.

Fixes: https://tracker.ceph.com/issues/50371

Signed-off-by: Brad Hubbard <bhubbard@redhat.com>
---

diff --git a/qa/clusters/2-node-mgr.yaml b/qa/clusters/2-node-mgr.yaml
index b1c29a866385..be657492c1ec 100644
--- a/qa/clusters/2-node-mgr.yaml
+++ b/qa/clusters/2-node-mgr.yaml
@@ -1,3 +1,4 @@
+clusternodes: 2
 roles:
 - [mgr.x, mon.a, mon.c, mds.a, mds.c, osd.0, client.0]
 - [mgr.y, mgr.z, mon.b, mds.b, osd.1, osd.2, osd.3, client.1]
diff --git a/qa/clusters/extra-client.yaml b/qa/clusters/extra-client.yaml
index 33fa505b7140..4bf258bcd419 100644
--- a/qa/clusters/extra-client.yaml
+++ b/qa/clusters/extra-client.yaml
@@ -1,3 +1,4 @@
+clusternodes: 2
 roles:
 - [mon.a, mon.c, osd.0, osd.1, osd.2]
 - [mon.b, mgr.x, mds.a, osd.3, osd.4, osd.5]
@@ -11,4 +12,4 @@ overrides:
   ceph:
     conf:
       osd:
-        osd shutdown pgref assert: true
\ No newline at end of file
+        osd shutdown pgref assert: true
diff --git a/qa/clusters/fixed-1.yaml b/qa/clusters/fixed-1.yaml
index d8e5898b99f4..a7cf5db3d69d 100644
--- a/qa/clusters/fixed-1.yaml
+++ b/qa/clusters/fixed-1.yaml
@@ -1,3 +1,4 @@
+clusternodes: 1
 overrides:
   ceph-deploy:
     conf:
diff --git a/qa/clusters/fixed-2.yaml b/qa/clusters/fixed-2.yaml
index e4448bb20088..964d6a49b2bf 100644
--- a/qa/clusters/fixed-2.yaml
+++ b/qa/clusters/fixed-2.yaml
@@ -1,3 +1,4 @@
+clusternodes: 2
 roles:
 - [mon.a, mon.c, mgr.y, osd.0, osd.1, osd.2, osd.3, client.0, node-exporter.a]
 - [mon.b, mgr.x, osd.4, osd.5, osd.6, osd.7, client.1, prometheus.a, node-exporter.b]
diff --git a/qa/clusters/fixed-3-cephfs.yaml b/qa/clusters/fixed-3-cephfs.yaml
index 9e021b3bd69b..493b044eb998 100644
--- a/qa/clusters/fixed-3-cephfs.yaml
+++ b/qa/clusters/fixed-3-cephfs.yaml
@@ -1,3 +1,4 @@
+clusternodes: 2
 roles:
 - [mon.a, mds.a, mgr.x, osd.0, osd.1]
 - [mon.b, mds.b, mon.c, mgr.y, osd.2, osd.3]
diff --git a/qa/clusters/fixed-3.yaml b/qa/clusters/fixed-3.yaml
index ddc79a84b60c..e87ae9392375 100644
--- a/qa/clusters/fixed-3.yaml
+++ b/qa/clusters/fixed-3.yaml
@@ -1,3 +1,4 @@
+clusternodes: 2
 roles:
 - [mon.a, mon.c, mgr.x, osd.0, osd.1, osd.2, osd.3]
 - [mon.b, mgr.y, osd.4, osd.5, osd.6, osd.7]
diff --git a/qa/clusters/fixed-4.yaml b/qa/clusters/fixed-4.yaml
index b73163751ab6..eef1d54fdd3f 100644
--- a/qa/clusters/fixed-4.yaml
+++ b/qa/clusters/fixed-4.yaml
@@ -1,3 +1,4 @@
+clusternodes: 4
 roles: 
 - [mon.a, mgr.y, osd.0, osd.4, osd.8, osd.12]
 - [mon.b, osd.1, osd.5, osd.9, osd.13] 
@@ -11,4 +12,4 @@ overrides:
   ceph:
     conf:
       osd:
-        osd shutdown pgref assert: true
\ No newline at end of file
+        osd shutdown pgref assert: true
diff --git a/qa/suites/rados/thrash/workloads/rados_api_tests.yaml b/qa/suites/rados/thrash/workloads/rados_api_tests.yaml
index 3e72897ae052..97520e6a8cc8 100644
--- a/qa/suites/rados/thrash/workloads/rados_api_tests.yaml
+++ b/qa/suites/rados/thrash/workloads/rados_api_tests.yaml
@@ -1,3 +1,7 @@
+teuthology:
+  postmerge:
+    # Don't run this test on tiny clusters
+    - if yaml.clusternodes < 4 then reject() end
 overrides:
   ceph:
     log-ignorelist:
@@ -16,6 +20,7 @@ overrides:
       osd:
         osd class load list: "*"
         osd class default list: "*"
+        osd client watch timeout: 240
 tasks:
 - workunit:
     clients:
diff --git a/qa/suites/rados/verify/tasks/rados_api_tests.yaml b/qa/suites/rados/verify/tasks/rados_api_tests.yaml
index e5a54e69e011..d23eda8cf10b 100644
--- a/qa/suites/rados/verify/tasks/rados_api_tests.yaml
+++ b/qa/suites/rados/verify/tasks/rados_api_tests.yaml
@@ -1,3 +1,7 @@
+teuthology:
+  postmerge:
+    # Don't run this test on tiny clusters
+    - if yaml.clusternodes < 4 then reject() end
 overrides:
   ceph:
     log-ignorelist:
@@ -23,7 +27,7 @@ overrides:
       osd:
         osd class load list: "*"
         osd class default list: "*"
-        osd client watch timeout: 120
+        osd client watch timeout: 240
 tasks:
 - workunit:
     timeout: 6h