feat: add maxScheduleCount to scheduler (#207)

Signed-off-by: Gaius <gaius.qi@gmail.com>
dragonflyoss · Nov 9, 2023 · ce4e751 · ce4e751
1 parent 1bc1ea4
commit ce4e751
Show file tree

Hide file tree

Showing 3 changed files with 22 additions and 18 deletions.
diff --git a/charts/dragonfly/Chart.yaml b/charts/dragonfly/Chart.yaml
@@ -3,8 +3,8 @@ name: dragonfly
 description: Dragonfly is an intelligent P2P based image and file distribution system
 icon: https://raw.githubusercontent.com/dragonflyoss/Dragonfly2/main/docs/images/logo/dragonfly.svg
 type: application
-version: 1.1.10
-appVersion: 2.1.21
+version: 1.1.11
+appVersion: 2.1.22
 keywords:
   - dragonfly
   - d7y
@@ -26,8 +26,8 @@ sources:
 
 annotations:
   artifacthub.io/changes: |
-    - Add maxProcs configuration to service.
-    - Update dragonfly image tag to v2.1.21.
+    - Update dragonfly image tag to v2.1.22.
+    - Add maxScheduleCount to scheduler.
 
   artifacthub.io/links: |
     - name: Chart Source
@@ -36,13 +36,13 @@ annotations:
       url: https://github.com/dragonflyoss/Dragonfly2
   artifacthub.io/images: |
     - name: manager
-      image: dragonflyoss/manager:v2.1.21
+      image: dragonflyoss/manager:v2.1.22
     - name: dfdaemon
-      image: dragonflyoss/dfdaemon:v2.1.21
+      image: dragonflyoss/dfdaemon:v2.1.22
     - name: scheduler
-      image: dragonflyoss/scheduler:v2.1.21
+      image: dragonflyoss/scheduler:v2.1.22
     - name: trainer
-      image: dragonflyoss/scheduler:v2.1.21
+      image: dragonflyoss/scheduler:v2.1.22
     - name: triton
       image: nvcr.io/nvidia/tritonserver:23.06-py3
 

diff --git a/charts/dragonfly/README.md b/charts/dragonfly/README.md
@@ -248,7 +248,7 @@ helm delete dragonfly --namespace dragonfly-system
 | dfdaemon.pullPolicy | string | `"IfNotPresent"` | Image pull policy. |
 | dfdaemon.pullSecrets | list | `[]` (defaults to global.imagePullSecrets). | Image pull secrets. |
 | dfdaemon.resources | object | `{"limits":{"cpu":"2","memory":"2Gi"},"requests":{"cpu":"0","memory":"0"}}` | Pod resource requests and limits. |
-| dfdaemon.tag | string | `"v2.1.21"` | Image tag. |
+| dfdaemon.tag | string | `"v2.1.22"` | Image tag. |
 | dfdaemon.terminationGracePeriodSeconds | string | `nil` | Pod terminationGracePeriodSeconds. |
 | dfdaemon.tolerations | list | `[]` | List of node taints to tolerate. |
 | externalManager.grpcPort | int | `65003` | External GRPC service port. |
@@ -359,7 +359,7 @@ helm delete dragonfly --namespace dragonfly-system
 | manager.service.annotations | object | `{}` | Service annotations. |
 | manager.service.labels | object | `{}` | Service labels. |
 | manager.service.type | string | `"ClusterIP"` | Service type. |
-| manager.tag | string | `"v2.1.21"` | Image tag. |
+| manager.tag | string | `"v2.1.22"` | Image tag. |
 | manager.terminationGracePeriodSeconds | string | `nil` | Pod terminationGracePeriodSeconds. |
 | manager.tolerations | list | `[]` | List of node taints to tolerate. |
 | mysql.auth.database | string | `"manager"` | Mysql database name. |
@@ -402,6 +402,7 @@ helm delete dragonfly --namespace dragonfly-system
 | scheduler.config.scheduler.gc.peerTTL | string | `"24h"` | peerTTL is the ttl of peer. If the peer has been downloaded by other peers, then PeerTTL will be reset. |
 | scheduler.config.scheduler.gc.pieceDownloadTimeout | string | `"30m"` | pieceDownloadTimeout is the timeout of downloading piece. |
 | scheduler.config.scheduler.gc.taskGCInterval | string | `"30m"` | taskGCInterval is the interval of task gc. If all the peers have been reclaimed in the task, then the task will also be reclaimed. |
+| scheduler.config.scheduler.maxScheduleCount | int | `30` | maxScheduleCount is max schedule count. If scheduling exceeds the maxScheduleCount, scheduler will notify peer back-to-source. |
 | scheduler.config.scheduler.retryBackToSourceLimit | int | `5` | retryBackToSourceLimit reaches the limit, then the peer back-to-source. |
 | scheduler.config.scheduler.retryInterval | string | `"50ms"` | Retry scheduling interval. |
 | scheduler.config.scheduler.retryLimit | int | `10` | Retry scheduling limit times. |
@@ -463,7 +464,7 @@ helm delete dragonfly --namespace dragonfly-system
 | scheduler.service.labels | object | `{}` | Service labels. |
 | scheduler.service.type | string | `"ClusterIP"` | Service type. |
 | scheduler.statefulsetAnnotations | object | `{}` | Statefulset annotations. |
-| scheduler.tag | string | `"v2.1.21"` | Image tag. |
+| scheduler.tag | string | `"v2.1.22"` | Image tag. |
 | scheduler.terminationGracePeriodSeconds | string | `nil` | Pod terminationGracePeriodSeconds. |
 | scheduler.tolerations | list | `[]` | List of node taints to tolerate. |
 | seedPeer.config.aliveTime | string | `"0s"` | Daemon alive time, when sets 0s, daemon will not auto exit, it is useful for longtime running. |
@@ -565,7 +566,7 @@ helm delete dragonfly --namespace dragonfly-system
 | seedPeer.replicas | int | `3` | Number of Pods to launch. |
 | seedPeer.resources | object | `{"limits":{"cpu":"2","memory":"4Gi"},"requests":{"cpu":"0","memory":"0"}}` | Pod resource requests and limits. |
 | seedPeer.statefulsetAnnotations | object | `{}` | Statefulset annotations. |
-| seedPeer.tag | string | `"v2.1.21"` | Image tag. |
+| seedPeer.tag | string | `"v2.1.22"` | Image tag. |
 | seedPeer.terminationGracePeriodSeconds | string | `nil` | Pod terminationGracePeriodSeconds. |
 | seedPeer.tolerations | list | `[]` | List of node taints to tolerate. |
 | trainer.config.console | bool | `false` | Console shows log on console. |
@@ -624,7 +625,7 @@ helm delete dragonfly --namespace dragonfly-system
 | trainer.service.annotations | object | `{}` | Service annotations. |
 | trainer.service.labels | object | `{}` | Service labels. |
 | trainer.service.type | string | `"ClusterIP"` | Service type. |
-| trainer.tag | string | `"v2.1.21"` | Image tag. |
+| trainer.tag | string | `"v2.1.22"` | Image tag. |
 | trainer.terminationGracePeriodSeconds | string | `nil` | Pod terminationGracePeriodSeconds. |
 | trainer.tolerations | list | `[]` | List of node taints to tolerate. |
 | triton.aws | object | `{"accessKeyID":"","region":"","secretAccessKey":""}` | Credentials information. |

diff --git a/charts/dragonfly/values.yaml b/charts/dragonfly/values.yaml
@@ -145,7 +145,7 @@ scheduler:
   # -- Image repository.
   image: dragonflyoss/scheduler
   # -- Image tag.
-  tag: v2.1.21
+  tag: v2.1.22
   # -- Image pull policy.
   pullPolicy: IfNotPresent
   # -- Image pull secrets.
@@ -229,6 +229,9 @@ scheduler:
       algorithm: default
       # -- backToSourceCount is single task allows the peer to back-to-source count.
       backToSourceCount: 3
+      # -- maxScheduleCount is max schedule count. If scheduling exceeds the maxScheduleCount,
+      # scheduler will notify peer back-to-source.
+      maxScheduleCount: 30
       # -- retryBackToSourceLimit reaches the limit, then the peer back-to-source.
       retryBackToSourceLimit: 5
       # -- Retry scheduling limit times.
@@ -518,7 +521,7 @@ seedPeer:
   # -- Image repository.
   image: dragonflyoss/dfdaemon
   # -- Image tag.
-  tag: v2.1.21
+  tag: v2.1.22
   # -- Image pull policy.
   pullPolicy: IfNotPresent
   # -- Image pull secrets.
@@ -864,7 +867,7 @@ dfdaemon:
   # -- Image repository.
   image: dragonflyoss/dfdaemon
   # -- Image tag.
-  tag: v2.1.21
+  tag: v2.1.22
   # -- Image pull policy.
   pullPolicy: IfNotPresent
   # -- Image pull secrets.
@@ -1204,7 +1207,7 @@ manager:
   # -- Image repository.
   image: dragonflyoss/manager
   # -- Image tag.
-  tag: v2.1.21
+  tag: v2.1.22
   # -- Image pull policy.
   pullPolicy: IfNotPresent
   # -- Image pull secrets.
@@ -1573,7 +1576,7 @@ trainer:
   # -- Image repository.
   image: dragonflyoss/trainer
   # -- Image tag.
-  tag: v2.1.21
+  tag: v2.1.22
   # -- Image pull policy.
   pullPolicy: IfNotPresent
   # -- Image pull secrets.