Fix bugs (#392)

* fix helm chart of exchange fix CI task Signed-off-by: ChenLong Ma <owlet42@126.com> * fix pulsar port of docker-compose Signed-off-by: ChenLong Ma <owlet42@126.com>
FederatedAI · Apr 2, 2021 · 0e4d113 · 0e4d113
1 parent 42bbf4f
commit 0e4d113
Show file tree

Hide file tree

Showing 9 changed files with 162 additions and 64 deletions.
diff --git a/build/ci/ansible/group_vars/all b/build/ci/ansible/group_vars/all
@@ -16,3 +16,5 @@ docker_registry: docker.io
 kube_create_timeout: 1800
 check_interval: 60
 fate_create_timeout: 1800
+retries: 3
+delay: 5
diff --git a/build/ci/ansible/roles/kubefate/tasks/main.yml b/build/ci/ansible/roles/kubefate/tasks/main.yml
@@ -12,61 +12,61 @@
 
     - name: Wait until lock file has been deleted or timeout
       block:
-      - name: Shell to Get Epoch
-        command: date +%s
-        register: ansible_date_time_epoch
-      - debug:
-          msg: "{{ ansible_date_time_epoch.stdout }}"
-
-      - name: Get info on lock_dir
-        stat:
-          path: "{{ lock_dir }}"
-        register: lock_dir_cfg
-      - debug:
-          msg: "{{ lock_dir_cfg.stat.mtime }}"
-
-      - name: Get Epochs Info
-        block:
-          - name: Module to Get Epoch
-            command: date +%s
-            register: ansible_date_time_epoch
-          - debug:
-              msg: "{{ ansible_date_time_epoch.stdout }}"
-
-          - name: Get info on lock_dir
-            stat:
-              path: "{{ lock_dir }}"
-            register: lock_dir_cfg
-          - debug:
-              msg: "{{ lock_dir_cfg.stat.mtime }}"
-
-          - name: Debug date comparison
-            debug:
-              msg: "wait_for timeout={{ (state_timeout - (ansible_date_time_epoch.stdout|float - lock_dir_cfg.stat.mtime))|int }}"
-
-          - name: wait for lock file to be absent
-            wait_for:
-              path: "{{ lock_dir }}"
-              state: absent
-              timeout: "{{ (state_timeout - (ansible_date_time_epoch.stdout|float - lock_dir_cfg.stat.mtime))|int }}"
-            ignore_errors: true
-            register: waitfor
-            when:
-              - (ansible_date_time_epoch.stdout|float - lock_dir_cfg.stat.mtime) > 0
-              - (ansible_date_time_epoch.stdout|float - lock_dir_cfg.stat.mtime) < state_timeout
-
-          - name: Remove lock file lock_dir
-            block:
-              - name: Remove lock file lock_dir
-                file:
-                  path: "{{ lock_dir }}"
-                  state: absent
-
-              - name: Sleep for {{ wait_processes_rm_lock_file }} seconds for all waiting processes to make the file absent, May need to increase this
-                wait_for:
-                  delay: "{{ wait_processes_rm_lock_file }}"
-                  timeout: 0
-            when: (waitfor.skipped is defined and waitfor.skipped) or (waitfor.state is defined and waitfor.state!="absent") 
+        - name: Shell to Get Epoch
+          command: date +%s
+          register: ansible_date_time_epoch
+        - debug:
+            msg: "{{ ansible_date_time_epoch.stdout }}"
+
+        - name: Get info on lock_dir
+          stat:
+            path: "{{ lock_dir }}"
+          register: lock_dir_cfg
+        - debug:
+            msg: "{{ lock_dir_cfg.stat.mtime }}"
+
+        - name: Get Epochs Info
+          block:
+            - name: Module to Get Epoch
+              command: date +%s
+              register: ansible_date_time_epoch
+            - debug:
+                msg: "{{ ansible_date_time_epoch.stdout }}"
+
+            - name: Get info on lock_dir
+              stat:
+                path: "{{ lock_dir }}"
+              register: lock_dir_cfg
+            - debug:
+                msg: "{{ lock_dir_cfg.stat.mtime }}"
+
+            - name: Debug date comparison
+              debug:
+                msg: "wait_for timeout={{ (state_timeout - (ansible_date_time_epoch.stdout|float - lock_dir_cfg.stat.mtime))|int }}"
+
+            - name: wait for lock file to be absent
+              wait_for:
+                path: "{{ lock_dir }}"
+                state: absent
+                timeout: "{{ (state_timeout - (ansible_date_time_epoch.stdout|float - lock_dir_cfg.stat.mtime))|int }}"
+              ignore_errors: true
+              register: waitfor
+              when:
+                - (ansible_date_time_epoch.stdout|float - lock_dir_cfg.stat.mtime) > 0
+                - (ansible_date_time_epoch.stdout|float - lock_dir_cfg.stat.mtime) < state_timeout
+
+            - name: Remove lock file lock_dir
+              block:
+                - name: Remove lock file lock_dir
+                  file:
+                    path: "{{ lock_dir }}"
+                    state: absent
+
+                - name: Sleep for {{ wait_processes_rm_lock_file }} seconds for all waiting processes to make the file absent, May need to increase this
+                  wait_for:
+                    delay: "{{ wait_processes_rm_lock_file }}"
+                    timeout: 0
+              when: (waitfor.skipped is defined and waitfor.skipped) or (waitfor.state is defined and waitfor.state!="absent")
       when: lockcode.rc|int==1
 
     - name: Start kubefate test
@@ -89,39 +89,55 @@
                   async: "{{ kube_create_timeout }}"
                   poll: "{{ check_interval }}"
                   shell: "bash {{ work_dir }}/utils/create_k8s.sh"
+                  retries: "{{ retries }}"
+                  delay: "{{ delay }}"
 
                 - name: Pull images
                   shell: "bash {{ work_dir }}/get_images.sh"
+                  retries: "{{ retries }}"
+                  delay: "{{ delay }}"
 
                 - name: kubefate build
                   async: "{{ kube_create_timeout }}"
                   poll: "{{ check_interval }}"
                   shell: "bash {{ work_dir }}/utils/kubefate_build.sh"
+                  retries: "{{ retries }}"
+                  delay: "{{ delay }}"
 
                 - name: deploy kubefate
                   async: "{{ kube_create_timeout }}"
                   poll: "{{ check_interval }}"
                   shell: "bash {{ work_dir }}/kubefate-e2e/kubefate_deploy_test.sh"
+                  retries: "{{ retries }}"
+                  delay: "{{ delay }}"
 
                 - name: check fate curd
                   async: "{{ kube_create_timeout }}"
                   poll: "{{ check_interval }}"
                   shell: "bash {{ work_dir }}/kubefate-e2e/fate_deploy_test.sh"
-
+                  retries: "{{ retries }}"
+                  delay: "{{ delay }}"
+
                 - name: Running deployment script.
                   async: "{{ fate_create_timeout }}"
                   poll: "{{ check_interval }}"
                   shell: "bash {{ work_dir }}/kubefate-e2e/deploy.sh"
+                  retries: "{{ retries }}"
+                  delay: "{{ delay }}"
 
               always:
                 - name: Clean up k8s environment
                   shell: "bash {{ work_dir }}/clean.sh"
+                  retries: "{{ retries }}"
+                  delay: "{{ delay }}"
 
           always:
             - name: Clean up the project files of the target machine
               file:
                 path: "{{ base_dir }}"
                 state: absent
+              retries: "{{ retries }}"
+              delay: "{{ delay }}"
 
       always:
         - name: Remove lock file lock_dir

diff --git a/docker-deploy/training_template/docker-compose-spark.yml b/docker-deploy/training_template/docker-compose-spark.yml
@@ -147,7 +147,7 @@ services:
     ports:
       - "6650:6650"
       - "6651:6651"
-      - "8000:8080"
+      - "8001:8080"
     deploy:
       restart_policy:
         condition: any

diff --git a/helm-charts/FATE-Exchange/partyId-values.yaml b/helm-charts/FATE-Exchange/partyId-values.yaml
@@ -9,14 +9,28 @@ image:
   imagePullSecrets: 
 #  - name: 
 
+exchange:
+  partyIp: 192.168.0.1
+  partyPort: 30000
+
+partyList:
+  - partyId: 10000
+    partyIp: 192.168.0.1
+    partyPort: 30101
+
 modules:
   rollsite: 
     include: false
+    type: NodePort
+    nodePort: 30091
+    partyList:
+      - partyId: 10000
+        partyIp: 192.168.0.1
+        partyPort: 30101
   trafficServer:
     include: true
     type: ClusterIP
-    httpNodePort: 30097
-    grpcNodePort: 30092
+    nodePort: 30097
     route_table: 
       sni:
       - fqdn: 10000.fate.org

diff --git a/helm-charts/FATE-Exchange/templates/_helpers.tpl b/helm-charts/FATE-Exchange/templates/_helpers.tpl
@@ -0,0 +1,16 @@
+{{/* Helm required labels */}}
+{{- define "fate.labels" -}}
+name: {{ .Values.partyName | quote  }}
+partyId: {{ .Values.partyId | quote  }}
+owner: kubefate
+cluster: fate-exchange
+heritage: {{ .Release.Service }}
+release: {{ .Release.Name }}
+chart: {{ .Chart.Name }}
+{{- end -}}
+
+{{/* matchLabels */}}
+{{- define "fate.matchLabels" -}}
+name: {{ .Values.partyName | quote  }}
+partyId: {{ .Values.partyId | quote  }}
+{{- end -}}
diff --git a/helm-charts/FATE-Exchange/templates/rollsite-module.yaml b/helm-charts/FATE-Exchange/templates/rollsite-module.yaml
@@ -18,6 +18,22 @@ metadata:
     fateMoudle: rollsite
 {{ include "fate.labels" . | indent 4 }}
 data:
+  eggroll.properties: |
+    # rollsite
+    eggroll.rollsite.coordinator={{ .Values.partyName }}
+    eggroll.rollsite.host=rollsite
+    eggroll.rollsite.port=9370
+    eggroll.rollsite.party.id={{ .Values.partyId }}
+    eggroll.rollsite.route.table.path=conf/route_table/route_table.json
+    eggroll.rollsite.route.table.key=
+    eggroll.rollsite.route.table.whitelist=127.0.0.1
+    eggroll.rollsite.jvm.options=-XX:+UseG1GC -XX:+PrintGCDetails -XX:+PrintGCDateStamps -Xloggc:logs/eggroll/rollsite.gc.log
+    
+    eggroll.rollsite.push.max.retry=3
+    eggroll.rollsite.push.long.retry=2
+    eggroll.rollsite.push.batches.per.stream=10
+    eggroll.rollsite.adapter.sendbuf.size=100000
+
   route_table.json: |
     {
         "route_table": {
@@ -115,7 +131,7 @@ spec:
             - mountPath: /data/projects/fate/eggroll/conf/route_table/
               name: rollsite-confs
             - mountPath: /data/projects/fate/eggroll/conf/eggroll.properties
-              name: eggroll-confs
+              name: rollsite-confs
               subPath: eggroll.properties
       {{- with .Values.modules.rollsite.nodeSelector }}
       nodeSelector: 
@@ -130,9 +146,6 @@ spec:
         - name: rollsite-confs
           configMap:
             name: rollsite-config
-        - name: eggroll-confs
-          configMap:
-            name: eggroll-config
 ---
 apiVersion: v1
 kind: Service

diff --git a/helm-charts/FATE-Exchange/values-template-example.yaml b/helm-charts/FATE-Exchange/values-template-example.yaml
@@ -12,6 +12,18 @@ istio:
 modules:
   - trafficServer
 
+rollsite: 
+  type: NodePort
+  nodePort: 30091
+  exchange:
+    ip: 192.168.0.1
+    port: 30000
+  partyList:
+  - partyId: 10000
+    partyIp: 192.168.0.1
+    partyPort: 30101
+  nodeSelector: 
+
 trafficServer:
   type: ClusterIP
   nodePort: 30097

diff --git a/helm-charts/FATE-Exchange/values-template.yaml b/helm-charts/FATE-Exchange/values-template.yaml
@@ -9,10 +9,27 @@ image:
   imagePullSecrets: 
 #  - name: 
 
+exchange:
+{{- with .rollsite }}
+{{- with .exchange }}
+  partyIp: {{ .ip }}
+  partyPort: {{ .port }}
+{{- end }}
+{{- end }}
+
+partyList:
+{{- with .rollsite }}
+{{- range .partyList }}
+  - partyId: {{ .partyId }}
+    partyIp: {{ .partyIp }}
+    partyPort: {{ .partyPort }}
+{{- end }}
+{{- end }}
+
 modules:
   rollsite: 
     include: {{ has "rollsite" .modules }}
-    {{- with .rabbitmq }}
+    {{- with .rollsite }}
     {{- with .nodeSelector }}
     nodeSelector: 
 {{ toYaml . | indent 6 }}

diff --git a/helm-charts/FATE-Exchange/values.yaml b/helm-charts/FATE-Exchange/values.yaml
@@ -9,6 +9,14 @@ image:
 partyId: 9999
 partyName: fate-9999
 
+partyList:
+- partyId: 8888
+  partyIp: 192.168.8.1
+  partyPort: 30081
+- partyId: 10000
+  partyIp: 192.168.10.1
+  partyPort: 30101
+
 modules:
   rollsite: 
     include: false