VGalaxies commented on code in PR #2676:
URL: https://github.com/apache/hugegraph/pull/2676#discussion_r3371421505
##########
hugegraph-server/hugegraph-core/src/main/java/org/apache/hugegraph/task/ServerInfoManager.java:
##########
@@ -228,6 +238,32 @@ protected boolean graphIsReady() {
return !this.closed && this.graph.started() &&
this.graph.initialized();
}
+ protected synchronized void updateIsSingleNode(){
+ Collection<HugeServerInfo> servers=this.allServerInfos();
+ boolean hasWorkerNode = false;
+ long now = DateUtil.now().getTime();
+ int computerNodeCount=0;
+
+ // Iterate servers to find suitable one
+ for (HugeServerInfo server : servers) {
+ if (!server.alive()) {
+ continue;
+ }
+ if (server.role().master()) {
+ continue;
+ }else if (server.role().computer()){
+ computerNodeCount++;
+ }
+ hasWorkerNode = true;
+ }
+
+ boolean singleNode = !hasWorkerNode||computerNodeCount==1;
Review Comment:
**High: One computer node marks every cluster as single-node**
`hugegraph-server/hugegraph-core/src/main/java/org/apache/hugegraph/task/ServerInfoManager.java:260`
**Evidence**
- `singleNode = !hasWorkerNode || computerNodeCount == 1` becomes true for
any live cluster with exactly one `COMPUTER`, even when live `WORKER` nodes
also exist. Then `selfIsMasterOrSingleComputer()` allows that computer node to
run `scheduleTasksOnMaster()` via `TaskManager.java:450-451`.
**Impact**
- A normal master/worker cluster with one computer node can now have a
non-master computer scheduling tasks concurrently with the real master, and
`onlySingleNode()` also makes the assigned master execute worker tasks locally
instead of returning in non-elected mode.
**Requested fix**
- Keep `onlySingleNode` as the original “no live non-master workers”
meaning, and add a separate standalone-computer predicate that is true only
when the current node is `COMPUTER` and the live server set contains no
master/worker and exactly that one computer.
##########
hugegraph-server/hugegraph-core/src/main/java/org/apache/hugegraph/task/ServerInfoManager.java:
##########
@@ -172,8 +172,18 @@ public NodeRole selfNodeRole() {
return this.globalNodeInfo.nodeRole();
}
- public boolean selfIsMaster() {
- return this.selfNodeRole() != null && this.selfNodeRole().master();
+ public boolean selfIsMasterOrSingleComputer() {
+ boolean isMaster=this.selfNodeRole() != null &&
this.selfNodeRole().master();
+ boolean isSingleComputer=isStandAloneComputer();
Review Comment:
**Medium: Standalone computer cannot schedule immediately after startup**
`hugegraph-server/hugegraph-core/src/main/java/org/apache/hugegraph/task/ServerInfoManager.java:177`
**Evidence**
- `onlySingleNode` is initialized false and `initServerInfo()` does not
recompute it; `selfIsMasterOrSingleComputer()` depends on that flag through
`isStandAloneComputer()`. The first recompute is `TaskManager.java:442`, but
the scheduler starts after `10 * SCHEDULE_PERIOD` at `TaskManager.java:103-106`.
**Impact**
- A single `COMPUTER` server still rejects task scheduling as “non-master”
during startup until the delayed background scheduler runs, making the feature
timing-dependent.
**Requested fix**
- Initialize/recompute standalone-computer state synchronously during
`initServerInfo()`, or make the master check derive standalone-computer status
from current live server info before rejecting scheduling.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
