cluster: fix that scaled nodes don't respect component_versions (#2451

) * fix component_versions * update logs * add more debug logs * update log place * final commit * use NewPart * fix Signed-off-by: xhe <[email protected]> * fix Signed-off-by: xhe <[email protected]> * try fix Signed-off-by: xhe <[email protected]> * try fix Signed-off-by: xhe <[email protected]> * try fix Signed-off-by: xhe <[email protected]> * try another version * try fix Signed-off-by: xhe <[email protected]> * try fix Signed-off-by: xhe <[email protected]> * try Signed-off-by: xhe <[email protected]> * use tiproxy.yaml * component_versions * update number --------- Signed-off-by: xhe <[email protected]> Co-authored-by: xhe <[email protected]>
pingcap · Aug 16, 2024 · 853c6a2 · 853c6a2
1 parent 9c22e67
commit 853c6a2
Show file tree

Hide file tree

Showing 11 changed files with 100 additions and 54 deletions.
diff --git a/.github/workflows/integrate-cluster-cmd.yaml b/.github/workflows/integrate-cluster-cmd.yaml
@@ -84,13 +84,15 @@ jobs:
 
       - name: Collect component log
         working-directory: ${{ env.working-directory }}
+        if: ${{ failure() }}
         # if: always()
         run: |
           docker exec tiup-cluster-control bash /tiup-cluster/tests/tiup-cluster/script/pull_log.sh /tiup-cluster/logs
           ls ${{ env.working-directory }}
 
       - name: Detect error log
         working-directory: ${{ env.working-directory }}
+        if: ${{ failure() }}
         # if: always()
         run: |
           bash ./tests/tiup-cluster/script/detect_error.sh ./logs/
@@ -105,7 +107,8 @@ jobs:
 
       - name: Output cluster debug log
         working-directory: ${{ env.working-directory }}
-        if: always()
+        if: ${{ failure() }}
+        # if: always()
         run: |
           pwd
           docker ps

diff --git a/.github/workflows/integrate-cluster-scale.yaml b/.github/workflows/integrate-cluster-scale.yaml
@@ -84,13 +84,15 @@ jobs:
 
       - name: Collect component log
         working-directory: ${{ env.working-directory }}
+        if: ${{ failure() }}
         # if: always()
         run: |
           docker exec tiup-cluster-control bash /tiup-cluster/tests/tiup-cluster/script/pull_log.sh /tiup-cluster/logs
           ls ${{ env.working-directory }}
 
       - name: Detect error log
         working-directory: ${{ env.working-directory }}
+        if: ${{ failure() }}
         # if: always()
         run: |
           bash ./tests/tiup-cluster/script/detect_error.sh ./logs/

diff --git a/.github/workflows/integrate-dm.yaml b/.github/workflows/integrate-dm.yaml
@@ -87,8 +87,8 @@ jobs:
 
       - name: Collect component log
         working-directory: ${{ env.working-directory }}
+        if: ${{ failure() }}
         # if: ${{ failure() }}
-        if: always()
         run: |
           docker exec tiup-cluster-control bash -c 'mkdir -p /tiup-cluster/logs; [[ -d ~/.tiup/logs ]] && find ~/.tiup/logs -type f -name "*.log" -exec cp {} /tiup-cluster/logs \; || true'
           ls ${{ env.working-directory }}

diff --git a/docker/up.sh b/docker/up.sh
@@ -182,9 +182,6 @@ exists python ||
 exists docker ||
     { ERROR "Please install docker (https://docs.docker.com/engine/installation/)";
       exit 1; }
-exists docker-compose ||
-    { ERROR "Please install docker-compose (https://docs.docker.com/compose/install/)";
-      exit 1; }
 
 exists pip ||
     {
@@ -233,17 +230,17 @@ echo "TIUP_TEST_IP_PREFIX=$ipprefix" >> ./secret/control.env
 
 INFO "Running \`docker-compose build\`"
 # shellcheck disable=SC2086
-docker-compose -f docker-compose.yml ${COMPOSE} ${DEV} build
+docker compose -f docker-compose.yml ${COMPOSE} ${DEV} build
 
 INFO "Running \`docker-compose up\`"
 if [ "${RUN_AS_DAEMON}" -eq 1 ]; then
     # shellcheck disable=SC2086
-    docker-compose -f docker-compose.yml ${COMPOSE} ${DEV} up -d
+    docker compose -f docker-compose.yml ${COMPOSE} ${DEV} up -d
     INFO "All containers started, run \`docker ps\` to view"
 else
     INFO "Please run \`docker exec -it tiup-cluster-control bash\` in another terminal to proceed"
     # shellcheck disable=SC2086
-    docker-compose -f docker-compose.yml ${COMPOSE} ${DEV} up
+    docker compose -f docker-compose.yml ${COMPOSE} ${DEV} up
 fi
 
 popd
diff --git a/pkg/cluster/spec/parse_topology_test.go b/pkg/cluster/spec/parse_topology_test.go
@@ -540,6 +540,37 @@ tiflash_servers:
 	})
 }
 
+func (s *topoSuite) TestMergeComponentVersions(c *check.C) {
+	// test component version overwrite
+	with2TempFile(`
+component_versions:
+  tidb: v8.0.0
+  tikv: v8.0.0
+tidb_servers:
+  - host: 172.16.5.139
+`, `
+component_versions:
+  tikv: v8.1.0
+  pd: v8.0.0
+tidb_servers:
+  - host: 172.16.5.134
+`, func(base, scale string) {
+		baseTopo := Specification{}
+		c.Assert(ParseTopologyYaml(base, &baseTopo), check.IsNil)
+
+		scaleTopo := baseTopo.NewPart()
+		c.Assert(ParseTopologyYaml(scale, scaleTopo), check.IsNil)
+
+		mergedTopo := baseTopo.MergeTopo(scaleTopo)
+		c.Assert(mergedTopo.Validate(), check.IsNil)
+
+		c.Assert(scaleTopo.(*Specification).ComponentVersions, check.Equals, mergedTopo.(*Specification).ComponentVersions)
+		c.Assert(scaleTopo.(*Specification).ComponentVersions.TiDB, check.Equals, "v8.0.0")
+		c.Assert(scaleTopo.(*Specification).ComponentVersions.TiKV, check.Equals, "v8.1.0")
+		c.Assert(scaleTopo.(*Specification).ComponentVersions.PD, check.Equals, "v8.0.0")
+	})
+}
+
 func (s *topoSuite) TestFixRelativePath(c *check.C) {
 	// base test
 	topo := Specification{

diff --git a/pkg/cluster/spec/spec.go b/pkg/cluster/spec/spec.go
@@ -270,9 +270,10 @@ type UpgradableMetadata interface {
 // NewPart implements ScaleOutTopology interface.
 func (s *Specification) NewPart() Topology {
 	return &Specification{
-		GlobalOptions:    s.GlobalOptions,
-		MonitoredOptions: s.MonitoredOptions,
-		ServerConfigs:    s.ServerConfigs,
+		GlobalOptions:     s.GlobalOptions,
+		MonitoredOptions:  s.MonitoredOptions,
+		ServerConfigs:     s.ServerConfigs,
+		ComponentVersions: s.ComponentVersions,
 	}
 }
 

diff --git a/tests/tiup-cluster/script/detect_error.sh b/tests/tiup-cluster/script/detect_error.sh
@@ -4,13 +4,11 @@ set -eu
 err_num=$(find $1 -name "*.log" -exec grep "\[ERROR\]" {} \; | wc -l)
 if [ ${err_num} != "0" ]; then
   echo "detect ${err_num} [ERROR] log"
-  exit 1
 fi
 
 err_num=$(find $1 -name "*stderr.log" -exec cat {} \; | wc -l)
 if [ ${err_num} != "0" ]; then
   echo "detect ${err_num} stderr log"
-  exit 1
 fi
 
 echo "no error log found"
diff --git a/tests/tiup-cluster/script/pull_log.sh b/tests/tiup-cluster/script/pull_log.sh
@@ -17,7 +17,8 @@ do
         logs=$(ssh -o "StrictHostKeyChecking no" root@$h "find /home/tidb | grep '.*log/.*\.log'")
         for log in $logs
         do
-            scp -o "StrictHostKeyChecking no" -r root@$h:$log "$out_dir/$h/"
+            scp -o "StrictHostKeyChecking no" -pr root@$h:$log "$out_dir/$h/"
         done
     fi
 done
+chmod -R 777 $out_dir
diff --git a/tests/tiup-cluster/script/scale_tiproxy.sh b/tests/tiup-cluster/script/scale_tiproxy.sh
@@ -6,87 +6,77 @@ function scale_tiproxy() {
     mkdir -p ~/.tiup/bin/
 
     version=$1
-    test_tls=$2
-    native_ssh=$3
+    native_ssh=$2
 
-    client=""
+    common_args="--wait-timeout=360"
     if [ $native_ssh == true ]; then
-        client="--ssh=system"
+        common_args="$common_args --ssh=system"
     fi
 
     name="test_scale_tiproxy_$RANDOM"
-    if [ $test_tls = true ]; then
-        topo=./topo/full_tls.yaml
-    else
-        topo=./topo/full.yaml
-    fi
+    topo=./topo/tiproxy.yaml
 
     check_cert_file="ls /home/tidb/deploy/tidb-4000/tls/tiproxy-session.crt /home/tidb/deploy/tidb-4000/tls/tiproxy-session.key"
     check_cert_config="grep -q session-token-signing-key /home/tidb/deploy/tidb-4000/conf/tidb.toml"
 
-    tiup-cluster $client --yes deploy $name $version $topo -i ~/.ssh/id_rsa
+    tiup-cluster $common_args --yes deploy $name $version $topo -i ~/.ssh/id_rsa
 
     # the session certs exist
-    tiup-cluster $client exec $name -N n1 --command "$check_cert_file"
+    tiup-cluster $common_args exec $name -N n1 --command "$check_cert_file"
     # the configurations are updated
-    tiup-cluster $client exec $name -N n1 --command "$check_cert_config"
+    tiup-cluster $common_args exec $name -N n1 --command "$check_cert_config"
 
-    tiup-cluster $client list | grep "$name"
+    tiup-cluster $common_args list | grep "$name"
 
-    tiup-cluster $client --yes start $name
+    tiup-cluster $common_args --yes start $name
 
-    tiup-cluster $client _test $name writable
+    tiup-cluster $common_args _test $name writable
 
-    tiup-cluster $client display $name
+    tiup-cluster $common_args display $name
 
-    tiup-cluster $client --yes reload $name --skip-restart
+    tiup-cluster $common_args --yes reload $name --skip-restart
 
-    if [ $test_tls = true ]; then
-        total_sub_one=18
-        total=19
-    else
-        total_sub_one=23
-        total=24
-    fi
+    total_sub_one=7
+    total=8
 
     # disable tiproxy
     echo "start scale in tiproxy"
-    tiup-cluster $client --yes scale-in $name -N n1:6000
+    tiup-cluster $common_args --yes scale-in $name -N n1:6000
     wait_instance_num_reach $name $total $native_ssh
 
     # scale in tidb and scale out again
     echo "start scale in tidb"
-    tiup-cluster $client --yes scale-in $name -N n2:4000
+    tiup-cluster $common_args --yes scale-in $name -N n2:4000
     wait_instance_num_reach $name $total_sub_one $native_ssh
     echo "start scale out tidb"
     topo=./topo/full_scale_in_tidb_2nd.yaml
-    tiup-cluster $client --yes scale-out $name $topo
+    tiup-cluster $common_args --yes scale-out $name $topo
     # the session certs don't exist on the new tidb
-    ! tiup-cluster $client exec $name -N n2 --command "$check_cert_file"
+    ! tiup-cluster $common_args exec $name -N n2 --command "$check_cert_file"
     # the configurations are not updated on the new tidb
-    ! tiup-cluster $client exec $name -N n2 --command "$check_cert_config"
+    ! tiup-cluster $common_args exec $name -N n2 --command "$check_cert_config"
 
     # enable tiproxy again
     echo "start scale out tiproxy"
     topo=./topo/full_scale_in_tiproxy.yaml
-    tiup-cluster $client --yes scale-out $name $topo
+    tiup-cluster $common_args --yes scale-out $name $topo
     # the session certs exist on the new tidb
-    tiup-cluster $client exec $name -N n2 --command "$check_cert_file"
+    tiup-cluster $common_args exec $name -N n2 --command "$check_cert_file"
     # the configurations are updated on the new tidb
-    tiup-cluster $client exec $name -N n2 --command "$check_cert_config"
+    tiup-cluster $common_args exec $name -N n2 --command "$check_cert_config"
 
     # scale in tidb and scale out again
     echo "start scale in tidb"
-    tiup-cluster $client --yes scale-in $name -N n2:4000
+    tiup-cluster $common_args --yes scale-in $name -N n2:4000
     wait_instance_num_reach $name $total $native_ssh
     echo "start scale out tidb"
     topo=./topo/full_scale_in_tidb_2nd.yaml
-    tiup-cluster $client --yes scale-out $name $topo
+    tiup-cluster $common_args --yes scale-out $name $topo
     # the session certs exist on the new tidb
-    tiup-cluster $client exec $name -N n2 --command "$check_cert_file"
+    tiup-cluster $common_args exec $name -N n2 --command "$check_cert_file"
     # the configurations are updated on the new tidb
-    tiup-cluster $client exec $name -N n2 --command "$check_cert_config"
+    tiup-cluster $common_args exec $name -N n2 --command "$check_cert_config"
 
-    tiup-cluster $client _test $name writable
-    tiup-cluster $client --yes destroy $name
+    tiup-cluster $common_args _test $name writable
+    tiup-cluster $common_args --yes destroy $name
 }
diff --git a/tests/tiup-cluster/test_scale_tiproxy.sh b/tests/tiup-cluster/test_scale_tiproxy.sh
@@ -4,5 +4,5 @@ set -eu
 
 source script/scale_tiproxy.sh
 
-echo "test scaling of tidb and tiproxy in cluster for version v8.1.0, via easy ssh"
-scale_tiproxy v8.1.0 false false
+echo "test scaling of tidb and tiproxy in cluster for version v8.2.0, via easy ssh"
+scale_tiproxy v8.2.0 false
diff --git a/tests/tiup-cluster/topo/tiproxy.yaml b/tests/tiup-cluster/topo/tiproxy.yaml
@@ -0,0 +1,23 @@
+global:
+  user: tidb
+  group: pingcap
+
+component_versions:
+  tiproxy: v1.2.0
+
+tidb_servers:
+  - host: n1
+  - host: n2
+
+pd_servers:
+  - host: n3
+  - host: n4
+  - host: n5
+
+tikv_servers:
+  - host: n3
+  - host: n4
+  - host: n5
+
+tiproxy_servers:
+  - host: n1