This is an automated email from the ASF dual-hosted git repository.
edcoleman pushed a commit to branch main
in repository https://gitbox.apache.org/repos/asf/accumulo.git
commit 46ed92329cf83911c5b43f348658941e0979f3f9
Merge: 8da73ce467 96b86a5f62
Author: Ed Coleman <edcole...@apache.org>
AuthorDate: Fri May 3 16:57:08 2024 +0000
Merge remote-tracking branch 'upstream/2.1'
includes:
- 96b86a5f62 - update for additional scan server metrics
- c488f788ad - adds scan server metrics
.../accumulo/core/metrics/MetricsProducer.java | 108 +++++++++++----------
.../org/apache/accumulo/tserver/ScanServer.java | 55 +++++++++--
.../apache/accumulo/tserver/ScanServerMetrics.java | 59 +++++++++++
.../accumulo/tserver/ThriftScanClientHandler.java | 4 +-
.../tserver/metrics/TabletServerScanMetrics.java | 24 +++--
.../apache/accumulo/tserver/ScanServerTest.java | 9 ++
.../apache/accumulo/test/metrics/MetricsIT.java | 11 ++-
7 files changed, 192 insertions(+), 78 deletions(-)
diff --cc
core/src/main/java/org/apache/accumulo/core/metrics/MetricsProducer.java
index 5e43b2a938,1bb2a1c10e..2fdf9172ab
--- a/core/src/main/java/org/apache/accumulo/core/metrics/MetricsProducer.java
+++ b/core/src/main/java/org/apache/accumulo/core/metrics/MetricsProducer.java
@@@ -599,9 -613,14 +600,9 @@@ public interface MetricsProducer
String METRICS_MINC_PREFIX = "accumulo.tserver.compactions.minc.";
String METRICS_MINC_QUEUED = METRICS_MINC_PREFIX + "queued";
String METRICS_MINC_RUNNING = METRICS_MINC_PREFIX + "running";
-
- String METRICS_REPLICATION_PREFIX = "accumulo.replication.";
- String METRICS_REPLICATION_QUEUE = METRICS_REPLICATION_PREFIX + "queue";
- String METRICS_REPLICATION_PENDING_FILES = METRICS_REPLICATION_PREFIX +
"files.pending";
- String METRICS_REPLICATION_PEERS = METRICS_REPLICATION_PREFIX + "peers";
- String METRICS_REPLICATION_THREADS = METRICS_REPLICATION_PREFIX + "threads";
+ String METRICS_MINC_PAUSED = METRICS_MINC_PREFIX + "paused";
- String METRICS_SCAN_PREFIX = "accumulo.tserver.scans.";
+ String METRICS_SCAN_PREFIX = "accumulo.scan.";
String METRICS_SCAN_TIMES = METRICS_SCAN_PREFIX + "times";
String METRICS_SCAN_OPEN_FILES = METRICS_SCAN_PREFIX + "files.open";
String METRICS_SCAN_RESULTS = METRICS_SCAN_PREFIX + "result";
@@@ -609,10 -628,15 +610,17 @@@
String METRICS_SCAN_START = METRICS_SCAN_PREFIX + "start";
String METRICS_SCAN_CONTINUE = METRICS_SCAN_PREFIX + "continue";
String METRICS_SCAN_CLOSE = METRICS_SCAN_PREFIX + "close";
- String METRICS_SCAN_BUSY_TIMEOUT = METRICS_SCAN_PREFIX + "busy.timeout";
+ String METRICS_SCAN_BUSY_TIMEOUT_COUNTER = METRICS_SCAN_PREFIX +
"busy.timeout.count";
+ String METRICS_SCAN_RESERVATION_TIMER = METRICS_SCAN_PREFIX +
"reservation.timer";
+ String METRICS_SCAN_QUERIES = METRICS_SCAN_PREFIX + "queries";
+ String METRICS_SCAN_QUERY_SCAN_RESULTS = METRICS_SCAN_PREFIX +
"query.results";
+ String METRICS_SCAN_QUERY_SCAN_RESULTS_BYTES = METRICS_SCAN_PREFIX +
"query.results.bytes";
+ String METRICS_SCAN_SCANNED_ENTRIES = METRICS_SCAN_PREFIX +
"query.scanned.entries";
+ String METRICS_SCAN_PAUSED_FOR_MEM = METRICS_SCAN_PREFIX +
"paused.for.memory";
+ String METRICS_SCAN_RETURN_FOR_MEM = METRICS_SCAN_PREFIX +
"return.early.for.memory";
+ String METRICS_SCAN_TABLET_METADATA_CACHE = METRICS_SCAN_PREFIX +
"tablet.metadata.cache";
+
String METRICS_TSERVER_PREFIX = "accumulo.tserver.";
String METRICS_TSERVER_ENTRIES = METRICS_TSERVER_PREFIX + "entries";
String METRICS_TSERVER_MEM_ENTRIES = METRICS_TSERVER_PREFIX + "entries.mem";
diff --cc
server/tserver/src/main/java/org/apache/accumulo/tserver/ScanServer.java
index ca4b28d06e,1e237e80e8..44ab680a98
--- a/server/tserver/src/main/java/org/apache/accumulo/tserver/ScanServer.java
+++ b/server/tserver/src/main/java/org/apache/accumulo/tserver/ScanServer.java
@@@ -76,17 -70,20 +76,18 @@@ import org.apache.accumulo.core.metadat
import org.apache.accumulo.core.metadata.StoredTabletFile;
import org.apache.accumulo.core.metadata.schema.Ample;
import org.apache.accumulo.core.metadata.schema.TabletMetadata;
+import org.apache.accumulo.core.metadata.schema.TabletsMetadata;
import org.apache.accumulo.core.metrics.MetricsInfo;
import org.apache.accumulo.core.securityImpl.thrift.TCredentials;
-import org.apache.accumulo.core.spi.scan.ScanServerSelector;
-import org.apache.accumulo.core.tabletserver.thrift.ActiveScan;
+import org.apache.accumulo.core.tabletscan.thrift.ActiveScan;
++import org.apache.accumulo.core.tabletscan.thrift.ScanServerBusyException;
+import org.apache.accumulo.core.tabletscan.thrift.TSampleNotPresentException;
+import org.apache.accumulo.core.tabletscan.thrift.TSamplerConfiguration;
+import org.apache.accumulo.core.tabletscan.thrift.TabletScanClientService;
+import org.apache.accumulo.core.tabletscan.thrift.TooManyFilesException;
import org.apache.accumulo.core.tabletserver.thrift.NoSuchScanIDException;
import org.apache.accumulo.core.tabletserver.thrift.NotServingTabletException;
-import org.apache.accumulo.core.tabletserver.thrift.ScanServerBusyException;
-import
org.apache.accumulo.core.tabletserver.thrift.TSampleNotPresentException;
-import org.apache.accumulo.core.tabletserver.thrift.TSamplerConfiguration;
-import org.apache.accumulo.core.tabletserver.thrift.TabletScanClientService;
-import org.apache.accumulo.core.tabletserver.thrift.TooManyFilesException;
-import org.apache.accumulo.core.trace.thrift.TInfo;
import org.apache.accumulo.core.util.Halt;
-import org.apache.accumulo.core.util.HostAndPort;
import org.apache.accumulo.core.util.UtilWaitThread;
import org.apache.accumulo.core.util.threads.ThreadPools;
import org.apache.accumulo.server.AbstractServer;
@@@ -125,8 -122,9 +126,10 @@@ import com.github.benmanes.caffeine.cac
import com.google.common.annotations.VisibleForTesting;
import com.google.common.base.Preconditions;
import com.google.common.collect.Sets;
+import com.google.common.net.HostAndPort;
+ import io.micrometer.core.instrument.Tag;
+
public class ScanServer extends AbstractServer
implements TabletScanClientService.Iface, TabletHostingServer {
@@@ -375,10 -376,12 +379,12 @@@
MetricsInfo metricsInfo = getContext().getMetricsInfo();
metricsInfo.addServiceTags(getApplicationName(), clientAddress);
+ metricsInfo.addCommonTags(List.of(Tag.of("resource.group", groupName)));
scanMetrics = new TabletServerScanMetrics();
+ scanServerMetrics = new ScanServerMetrics(tabletMetadataCache);
- metricsInfo.addMetricsProducers(this, scanMetrics);
- metricsInfo.addMetricsProducers(scanMetrics, scanServerMetrics);
++ metricsInfo.addMetricsProducers(this, scanMetrics, scanServerMetrics);
metricsInfo.init();
// We need to set the compaction manager so that we don't get an NPE in
CompactableImpl.close
diff --cc
server/tserver/src/main/java/org/apache/accumulo/tserver/metrics/TabletServerScanMetrics.java
index 9a1faa6261,8e066dd7f7..09f8431bb4
---
a/server/tserver/src/main/java/org/apache/accumulo/tserver/metrics/TabletServerScanMetrics.java
+++
b/server/tserver/src/main/java/org/apache/accumulo/tserver/metrics/TabletServerScanMetrics.java
@@@ -39,9 -39,7 +39,9 @@@ public class TabletServerScanMetrics im
private Counter startScanCalls;
private Counter continueScanCalls;
private Counter closeScanCalls;
- private Counter busyTimeoutReturned;
+ private Counter busyTimeoutCount;
+ private Counter pausedForMemory;
+ private Counter earlyReturnForMemory;
private final LongAdder lookupCount = new LongAdder();
private final LongAdder queryResultCount = new LongAdder();
@@@ -72,10 -70,10 +72,6 @@@
return this.queryResultBytes.sum();
}
-- public void incrementScannedCount(long amount) {
-- this.scannedCount.add(amount);
-- }
--
public LongAdder getScannedCounter() {
return this.scannedCount;
}
@@@ -116,18 -114,10 +112,18 @@@
closeScanCalls.increment(value);
}
- public void incrementScanBusyTimeout(double value) {
- busyTimeoutReturned.increment(value);
+ public void incrementBusy(double value) {
+ busyTimeoutCount.increment(value);
}
+ public void incrementScanPausedForLowMemory() {
+ pausedForMemory.increment();
+ }
+
+ public void incrementEarlyReturnForLowMemory() {
+ earlyReturnForMemory.increment();
+ }
+
@Override
public void registerMetrics(MeterRegistry registry) {
Gauge.builder(METRICS_SCAN_OPEN_FILES, openFiles::get)
@@@ -143,23 -133,20 +139,25 @@@
.description("calls to continue a scan /
multiscan").register(registry);
closeScanCalls = Counter.builder(METRICS_SCAN_CLOSE)
.description("calls to close a scan / multiscan").register(registry);
- busyTimeoutReturned = Counter.builder(METRICS_SCAN_BUSY_TIMEOUT)
- .description("times that a scan has timed out in the
queue").register(registry);
- Gauge.builder(METRICS_TSERVER_QUERIES, this,
TabletServerScanMetrics::getLookupCount)
+ busyTimeoutCount = Counter.builder(METRICS_SCAN_BUSY_TIMEOUT_COUNTER)
+ .description("The number of scans where a busy timeout
happened").register(registry);
+ Gauge.builder(METRICS_SCAN_QUERIES, this,
TabletServerScanMetrics::getLookupCount)
.description("Number of queries").register(registry);
- Gauge.builder(METRICS_TSERVER_SCAN_RESULTS, this,
TabletServerScanMetrics::getQueryResultCount)
+ Gauge
+ .builder(METRICS_SCAN_QUERY_SCAN_RESULTS, this,
+ TabletServerScanMetrics::getQueryResultCount)
.description("Query rate (entries/sec)").register(registry);
Gauge
- .builder(METRICS_TSERVER_SCAN_RESULTS_BYTES, this,
+ .builder(METRICS_SCAN_QUERY_SCAN_RESULTS_BYTES, this,
TabletServerScanMetrics::getQueryByteCount)
.description("Query rate (bytes/sec)").register(registry);
- Gauge.builder(METRICS_TSERVER_SCANNED_ENTRIES, this,
TabletServerScanMetrics::getScannedCount)
+ Gauge.builder(METRICS_SCAN_SCANNED_ENTRIES, this,
TabletServerScanMetrics::getScannedCount)
.description("Scanned rate").register(registry);
+ pausedForMemory = Counter.builder(METRICS_SCAN_PAUSED_FOR_MEM)
+ .description("scan paused due to server being low on
memory").register(registry);
+ earlyReturnForMemory = Counter.builder(METRICS_SCAN_RETURN_FOR_MEM)
+ .description("scan returned results early due to server being low on
memory")
+ .register(registry);
}
}
diff --cc test/src/main/java/org/apache/accumulo/test/metrics/MetricsIT.java
index 4b92480f19,55622d0793..f2be6e71a5
--- a/test/src/main/java/org/apache/accumulo/test/metrics/MetricsIT.java
+++ b/test/src/main/java/org/apache/accumulo/test/metrics/MetricsIT.java
@@@ -99,10 -99,12 +99,13 @@@ public class MetricsIT extends Configur
doWorkToGenerateMetrics();
cluster.stop();
-- Set<String> unexpectedMetrics = Set.of(METRICS_SCAN_YIELDS,
METRICS_UPDATE_ERRORS,
- METRICS_COMPACTOR_MAJC_STUCK, METRICS_SCAN_BUSY_TIMEOUT,
METRICS_SCAN_PAUSED_FOR_MEM,
- METRICS_SCAN_RETURN_FOR_MEM, METRICS_MINC_PAUSED,
METRICS_MAJC_PAUSED);
- Set<String> flakyMetrics = Set.of(METRICS_GC_WAL_ERRORS,
METRICS_FATE_TYPE_IN_PROGRESS);
- METRICS_REPLICATION_QUEUE, METRICS_COMPACTOR_MAJC_STUCK,
METRICS_SCAN_BUSY_TIMEOUT_COUNTER);
- // add sserver as flaky until scan server included in mini tests.
++ Set<String> unexpectedMetrics =
++ Set.of(METRICS_SCAN_YIELDS, METRICS_UPDATE_ERRORS,
METRICS_COMPACTOR_MAJC_STUCK,
++ METRICS_SCAN_BUSY_TIMEOUT_COUNTER, METRICS_SCAN_PAUSED_FOR_MEM,
++ METRICS_SCAN_RETURN_FOR_MEM, METRICS_MINC_PAUSED,
METRICS_MAJC_PAUSED);
+ Set<String> flakyMetrics = Set.of(METRICS_GC_WAL_ERRORS,
METRICS_FATE_TYPE_IN_PROGRESS,
+ METRICS_SCAN_BUSY_TIMEOUT_COUNTER, METRICS_SCAN_RESERVATION_TIMER,
+ METRICS_SCAN_TABLET_METADATA_CACHE);
Map<String,String> expectedMetricNames = this.getMetricFields();
flakyMetrics.forEach(expectedMetricNames::remove); // might not see these
