Merge pull request PelicanPlatform#1677 from bbockelm/valgrind_tuning

Tuning Pelican for running xrootd under valgrind
jhiemstrawisc · Nov 27, 2024 · 8e5f60a · 8e5f60a
2 parents ee6adbe + 62a2854
commit 8e5f60a
Show file tree

Hide file tree

Showing 6 changed files with 33 additions and 3 deletions.
diff --git a/config/resources/defaults.yaml b/config/resources/defaults.yaml
@@ -60,6 +60,7 @@ Director:
   CachePresenceTTL: 1m
   CachePresenceCapacity: 10000
 Cache:
+  DefaultCacheTimeout: "9.5s"
   Port: 8442
   SelfTest: true
   SelfTestInterval: 15s
@@ -100,6 +101,7 @@ Shoveler:
   PortHigher: 9999
   AMQPExchange: shoveled-xrd
 Xrootd:
+  MaxStartupWait: "10s"
   Mount: ""
   ManagerPort: 1213
   DetailedMonitoringPort: 9930

diff --git a/docs/parameters.yaml b/docs/parameters.yaml
@@ -1326,6 +1326,16 @@ type: int
 default: 0
 components: ["cache"]
 ---
+name: Cache.DefaultCacheTimeout
+description: |+
+  The default value of the cache operation timeout if one is not specified by the client.
+
+  Newer clients should always specify a timeout; changing this default is rarely necessary.
+type: duration
+default: 9.5s
+hidden: true
+components: ["cache"]
+---
 ############################
 #  Director-level configs  #
 ############################
@@ -2355,6 +2365,15 @@ type: string
 default: none
 components: ["origin"]
 ---
+name: Xrootd.MaxStartupWait
+description: |+
+  The maximum amount of time pelican will wait for the xrootd daemons to
+  successfully start
+type: duration
+default: 10s
+hidden: true
+components: ["origin", "cache"]
+---
 ############################
 # Monitoring-level configs #
 ############################

diff --git a/images/dev.Dockerfile b/images/dev.Dockerfile
@@ -56,6 +56,7 @@ RUN yum install -y --enablerepo=osg-testing goreleaser npm xrootd-devel xrootd-s
     xrdcl-http jq procps docker make curl-devel java-17-openjdk-headless git cmake3 gcc-c++ openssl-devel sqlite-devel libcap-devel sssd-client \
     xrootd-multiuser \
     zlib-devel \
+    vim valgrind gdb gtest-devel \
     && yum clean all
 
 # The ADD command with a api.github.com URL in the next couple of sections

diff --git a/param/parameters.go b/param/parameters.go
diff --git a/param/parameters_struct.go b/param/parameters_struct.go
diff --git a/xrootd/launch.go b/xrootd/launch.go
@@ -91,6 +91,8 @@ func makeUnprivilegedXrootdLauncher(daemonName string, configPath string, isCach
 		if confDir := os.Getenv("XRD_PLUGINCONFDIR"); confDir != "" {
 			result.ExtraEnv = append(result.ExtraEnv, "XRD_PLUGINCONFDIR="+confDir)
 		}
+		result.ExtraEnv = append(result.ExtraEnv, "XRD_PELICANFEDERATIONMETADATATIMEOUT="+param.Cache_DefaultCacheTimeout.GetDuration().String())
+		result.ExtraEnv = append(result.ExtraEnv, "XRD_PELICANDEFAULTHEADERTIMEOUT="+param.Cache_DefaultCacheTimeout.GetDuration().String())
 	}
 	return
 }
@@ -165,7 +167,7 @@ func LaunchDaemons(ctx context.Context, launchers []daemon.Launcher, egrp *errgr
 		return
 	}
 
-	ticker := time.NewTicker(10 * time.Second)
+	ticker := time.NewTicker(param.Xrootd_MaxStartupWait.GetDuration())
 	defer ticker.Stop()
 	select {
 	case <-ctx.Done():
@@ -180,8 +182,8 @@ func LaunchDaemons(ctx context.Context, launchers []daemon.Launcher, egrp *errgr
 			portStartCallback(port)
 		}
 	case <-ticker.C:
-		log.Errorln("XRootD did not startup after 10s of waiting")
-		err = errors.New("XRootD did not startup after 10s of waiting")
+		log.Errorln("XRootD did not startup after", param.Xrootd_MaxStartupWait.GetDuration().String(), "of waiting")
+		err = errors.New("XRootD did not startup after " + param.Xrootd_MaxStartupWait.GetDuration().String() + " of waiting")
 		return
 	}