mqliang commented on a change in pull request #7368:
URL: https://github.com/apache/pinot/pull/7368#discussion_r718894821
##########
File path:
pinot-plugins/pinot-minion-tasks/pinot-minion-builtin-tasks/src/main/java/org/apache/pinot/plugin/minion/tasks/mergerollup/MergeRollupTaskGenerator.java
##########
@@ -463,4 +490,71 @@ private long getWatermarkMs(long minStartTimeMs, long
bucketMs, String mergeLeve
return pinotTaskConfigs;
}
+
+ private long getMergeRollupTaskDelayInNumTimeBuckets(long watermarkMs, long
bufferTimeMs, long bucketTimeMs) {
+ if (bufferTimeMs == 0 || watermarkMs == -1) {
+ return 0;
+ }
+ return (long) Math.floor((System.currentTimeMillis() - watermarkMs -
bufferTimeMs) / (double) bucketTimeMs);
+ }
+
+ private void setWatermarkMs(String tableNameWithType, String mergeLevel,
long watermarkMs, long bufferTimeMs,
+ long bucketTimeMs) {
+ LOGGER.info(
+ "Setting watermark for table: {} and mergeLevel: {} is {}
(watermarkMs={}, bufferTimeMs={}, bucketTimeMs={})",
+ tableNameWithType, mergeLevel,
getMergeRollupTaskDelayInNumTimeBuckets(watermarkMs, bufferTimeMs,
bucketTimeMs),
+ watermarkMs, bucketTimeMs, bucketTimeMs);
+
+ ControllerMetrics controllerMetrics =
_clusterInfoAccessor.getControllerMetrics();
+ if (controllerMetrics == null) {
+ return;
+ }
+
+ PinotMetricsRegistry metricsRegistry =
controllerMetrics.getMetricsRegistry();
+ if (metricsRegistry == null) {
+ return;
+ }
+
+ // Update gauge value that indicates the delay in terms of the number of
time buckets.
+ _mergeRollupWatermarks.computeIfAbsent(tableNameWithType, k -> new
HashMap<>());
+ Map<String, Long> watermarkForTable =
_mergeRollupWatermarks.get(tableNameWithType);
+
+ watermarkForTable.compute(mergeLevel, (k, v) -> {
+ if (v == null) {
+
controllerMetrics.addCallbackGaugeIfNeeded(getMetricNameForTaskDelay(tableNameWithType,
mergeLevel),
+ (() ->
getMergeRollupTaskDelayInNumTimeBuckets(watermarkForTable.getOrDefault(k, -1L),
bufferTimeMs,
+ bucketTimeMs)));
+ }
Review comment:
Do we need add a `else` block here which remove the old CallbackGauge
and add a new CallbackGauge when `v != null`?
##########
File path:
pinot-plugins/pinot-minion-tasks/pinot-minion-builtin-tasks/src/main/java/org/apache/pinot/plugin/minion/tasks/mergerollup/MergeRollupTaskGenerator.java
##########
@@ -463,4 +490,71 @@ private long getWatermarkMs(long minStartTimeMs, long
bucketMs, String mergeLeve
return pinotTaskConfigs;
}
+
+ private long getMergeRollupTaskDelayInNumTimeBuckets(long watermarkMs, long
bufferTimeMs, long bucketTimeMs) {
+ if (bufferTimeMs == 0 || watermarkMs == -1) {
+ return 0;
+ }
+ return (long) Math.floor((System.currentTimeMillis() - watermarkMs -
bufferTimeMs) / (double) bucketTimeMs);
Review comment:
Is it possible that `bucketTimeMs ` be 0? which will cause "divide by
zero" exception.
##########
File path:
pinot-plugins/pinot-minion-tasks/pinot-minion-builtin-tasks/src/main/java/org/apache/pinot/plugin/minion/tasks/mergerollup/MergeRollupTaskGenerator.java
##########
@@ -339,6 +351,21 @@ public String getTaskType() {
.info("Finished generating task configs for table: {} for task: {},
numTasks: {}", offlineTableName, taskType,
pinotTaskConfigsForTable.size());
}
+
+ // Reset watermarks for invalid tables. This covers the metrics clean up
when the table is removed or the merge
+ // config is added and then removed.
+ LeadControllerManager leadControllerManager =
_clusterInfoAccessor.getLeaderControllerManager();
Review comment:
L385-L268 can be changed as
```
for (String tableNameWithType : _mergeRollupWatermarks.keySet()) {
if (!leadControllerManager.isLeaderForTable(tableNameWithType)) {
resetWatermarkMs(tableNameWithType);
}
}
```
Since _mergeRollupWatermarks is a HashMap, so looping over
_mergeRollupWatermarks.keySet() will not iterating duplicate keys, so the
`!candidateMergeTables.contains(tableNameWithType)` check is not needed.
##########
File path:
pinot-plugins/pinot-minion-tasks/pinot-minion-builtin-tasks/src/main/java/org/apache/pinot/plugin/minion/tasks/mergerollup/MergeRollupTaskGenerator.java
##########
@@ -339,6 +351,21 @@ public String getTaskType() {
.info("Finished generating task configs for table: {} for task: {},
numTasks: {}", offlineTableName, taskType,
pinotTaskConfigsForTable.size());
}
+
+ // Reset watermarks for invalid tables. This covers the metrics clean up
when the table is removed or the merge
+ // config is added and then removed.
+ LeadControllerManager leadControllerManager =
_clusterInfoAccessor.getLeaderControllerManager();
Review comment:
L385-L268 can be changed as
```
for (String tableNameWithType : _mergeRollupWatermarks.keySet()) {
if (!leadControllerManager.isLeaderForTable(tableNameWithType)) {
resetWatermarkMs(tableNameWithType);
}
}
```
Since _mergeRollupWatermarks is a HashMap, looping over
_mergeRollupWatermarks.keySet() will not iterating duplicate keys, so the
`!candidateMergeTables.contains(tableNameWithType)` check is not needed.
##########
File path:
pinot-plugins/pinot-minion-tasks/pinot-minion-builtin-tasks/src/main/java/org/apache/pinot/plugin/minion/tasks/mergerollup/MergeRollupTaskGenerator.java
##########
@@ -339,6 +351,21 @@ public String getTaskType() {
.info("Finished generating task configs for table: {} for task: {},
numTasks: {}", offlineTableName, taskType,
pinotTaskConfigsForTable.size());
}
+
+ // Reset watermarks for invalid tables. This covers the metrics clean up
when the table is removed or the merge
+ // config is added and then removed.
+ LeadControllerManager leadControllerManager =
_clusterInfoAccessor.getLeaderControllerManager();
Review comment:
L385-L268 can be changed as
```
for (String tableNameWithType : _mergeRollupWatermarks.keySet()) {
if (!leadControllerManager.isLeaderForTable(tableNameWithType)) {
resetWatermarkMs(tableNameWithType);
}
}
```
Since _mergeRollupWatermarks is a HashMap, looping over
_mergeRollupWatermarks.keySet() will not produce duplicate keys, so the
`!candidateMergeTables.contains(tableNameWithType)` check is not needed.
##########
File path:
pinot-plugins/pinot-minion-tasks/pinot-minion-builtin-tasks/src/main/java/org/apache/pinot/plugin/minion/tasks/mergerollup/MergeRollupTaskGenerator.java
##########
@@ -463,4 +490,71 @@ private long getWatermarkMs(long minStartTimeMs, long
bucketMs, String mergeLeve
return pinotTaskConfigs;
}
+
+ private long getMergeRollupTaskDelayInNumTimeBuckets(long watermarkMs, long
bufferTimeMs, long bucketTimeMs) {
+ if (bufferTimeMs == 0 || watermarkMs == -1) {
+ return 0;
+ }
+ return (long) Math.floor((System.currentTimeMillis() - watermarkMs -
bufferTimeMs) / (double) bucketTimeMs);
+ }
+
+ private void setWatermarkMs(String tableNameWithType, String mergeLevel,
long watermarkMs, long bufferTimeMs,
+ long bucketTimeMs) {
+ LOGGER.info(
+ "Setting watermark for table: {} and mergeLevel: {} is {}
(watermarkMs={}, bufferTimeMs={}, bucketTimeMs={})",
+ tableNameWithType, mergeLevel,
getMergeRollupTaskDelayInNumTimeBuckets(watermarkMs, bufferTimeMs,
bucketTimeMs),
+ watermarkMs, bucketTimeMs, bucketTimeMs);
+
+ ControllerMetrics controllerMetrics =
_clusterInfoAccessor.getControllerMetrics();
+ if (controllerMetrics == null) {
+ return;
+ }
+
+ PinotMetricsRegistry metricsRegistry =
controllerMetrics.getMetricsRegistry();
+ if (metricsRegistry == null) {
+ return;
+ }
+
+ // Update gauge value that indicates the delay in terms of the number of
time buckets.
+ _mergeRollupWatermarks.computeIfAbsent(tableNameWithType, k -> new
HashMap<>());
+ Map<String, Long> watermarkForTable =
_mergeRollupWatermarks.get(tableNameWithType);
+
+ watermarkForTable.compute(mergeLevel, (k, v) -> {
+ if (v == null) {
+
controllerMetrics.addCallbackGaugeIfNeeded(getMetricNameForTaskDelay(tableNameWithType,
mergeLevel),
+ (() ->
getMergeRollupTaskDelayInNumTimeBuckets(watermarkForTable.getOrDefault(k, -1L),
bufferTimeMs,
+ bucketTimeMs)));
+ }
+ return watermarkMs;
+ });
+ }
+
+ /**
+ * Reset the watermark for the given table name
+ * @param tableNameWithType a table name with type
+ */
+ private void resetWatermarkMs(String tableNameWithType) {
+ ControllerMetrics controllerMetrics =
_clusterInfoAccessor.getControllerMetrics();
+ if (controllerMetrics == null) {
+ return;
+ }
+ PinotMetricsRegistry metricsRegistry =
controllerMetrics.getMetricsRegistry();
+ if (metricsRegistry == null) {
+ return;
+ }
+
+ // Delete all the watermarks associated with the given table name
+ Map<String, Long> watermarksForTable =
_mergeRollupWatermarks.get(tableNameWithType);
Review comment:
> (nit) Directly call _mergeRollupWatermarks.remove(tableNameWithType);
The current logic is correct IMO. As we need remove the CallbackGauge from
controllerMetrics also. Directly calling `
_mergeRollupWatermarks.remove(tableNameWithType);` only remove the
watermarksForTable map from `_mergeRollupWatermarks` , but the CallbackGauge is
still there.
##########
File path:
pinot-plugins/pinot-minion-tasks/pinot-minion-builtin-tasks/src/main/java/org/apache/pinot/plugin/minion/tasks/mergerollup/MergeRollupTaskGenerator.java
##########
@@ -463,4 +490,71 @@ private long getWatermarkMs(long minStartTimeMs, long
bucketMs, String mergeLeve
return pinotTaskConfigs;
}
+
+ private long getMergeRollupTaskDelayInNumTimeBuckets(long watermarkMs, long
bufferTimeMs, long bucketTimeMs) {
+ if (bufferTimeMs == 0 || watermarkMs == -1) {
+ return 0;
+ }
+ return (long) Math.floor((System.currentTimeMillis() - watermarkMs -
bufferTimeMs) / (double) bucketTimeMs);
+ }
+
+ private void setWatermarkMs(String tableNameWithType, String mergeLevel,
long watermarkMs, long bufferTimeMs,
+ long bucketTimeMs) {
+ LOGGER.info(
+ "Setting watermark for table: {} and mergeLevel: {} is {}
(watermarkMs={}, bufferTimeMs={}, bucketTimeMs={})",
+ tableNameWithType, mergeLevel,
getMergeRollupTaskDelayInNumTimeBuckets(watermarkMs, bufferTimeMs,
bucketTimeMs),
+ watermarkMs, bucketTimeMs, bucketTimeMs);
+
+ ControllerMetrics controllerMetrics =
_clusterInfoAccessor.getControllerMetrics();
+ if (controllerMetrics == null) {
+ return;
+ }
+
+ PinotMetricsRegistry metricsRegistry =
controllerMetrics.getMetricsRegistry();
+ if (metricsRegistry == null) {
+ return;
+ }
+
+ // Update gauge value that indicates the delay in terms of the number of
time buckets.
+ _mergeRollupWatermarks.computeIfAbsent(tableNameWithType, k -> new
HashMap<>());
+ Map<String, Long> watermarkForTable =
_mergeRollupWatermarks.get(tableNameWithType);
+
+ watermarkForTable.compute(mergeLevel, (k, v) -> {
+ if (v == null) {
+
controllerMetrics.addCallbackGaugeIfNeeded(getMetricNameForTaskDelay(tableNameWithType,
mergeLevel),
+ (() ->
getMergeRollupTaskDelayInNumTimeBuckets(watermarkForTable.getOrDefault(k, -1L),
bufferTimeMs,
+ bucketTimeMs)));
+ }
+ return watermarkMs;
+ });
+ }
+
+ /**
+ * Reset the watermark for the given table name
+ * @param tableNameWithType a table name with type
+ */
+ private void resetWatermarkMs(String tableNameWithType) {
+ ControllerMetrics controllerMetrics =
_clusterInfoAccessor.getControllerMetrics();
+ if (controllerMetrics == null) {
+ return;
+ }
+ PinotMetricsRegistry metricsRegistry =
controllerMetrics.getMetricsRegistry();
+ if (metricsRegistry == null) {
+ return;
+ }
+
+ // Delete all the watermarks associated with the given table name
+ Map<String, Long> watermarksForTable =
_mergeRollupWatermarks.get(tableNameWithType);
Review comment:
> (nit) Directly call _mergeRollupWatermarks.remove(tableNameWithType);
The current logic is correct IIUC. As we need remove the CallbackGauge from
controllerMetrics also. Directly calling `
_mergeRollupWatermarks.remove(tableNameWithType);` only remove the
watermarksForTable map from `_mergeRollupWatermarks` , but the CallbackGauge is
still there.
--
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]
