diff options
author | Martin Polden <mpolden@mpolden.no> | 2021-01-07 13:54:05 +0100 |
---|---|---|
committer | Martin Polden <mpolden@mpolden.no> | 2021-01-07 13:54:05 +0100 |
commit | 672d9a71ebac45a886c9ee45625890f18af8aa4f (patch) | |
tree | 1671b6dbe9554a9b743400476650bacd1ff60790 /vespajlib/src/main/java/com/yahoo | |
parent | 5c1096afe9f2db748f6e8d3c559021101fd8ae6c (diff) |
Count lock timeout as unsuccessful run for exclusive maintainers
Diffstat (limited to 'vespajlib/src/main/java/com/yahoo')
-rw-r--r-- | vespajlib/src/main/java/com/yahoo/concurrent/maintenance/JobMetrics.java | 3 | ||||
-rw-r--r-- | vespajlib/src/main/java/com/yahoo/concurrent/maintenance/Maintainer.java | 36 |
2 files changed, 25 insertions, 14 deletions
diff --git a/vespajlib/src/main/java/com/yahoo/concurrent/maintenance/JobMetrics.java b/vespajlib/src/main/java/com/yahoo/concurrent/maintenance/JobMetrics.java index 483057a828d..d4d60723cbe 100644 --- a/vespajlib/src/main/java/com/yahoo/concurrent/maintenance/JobMetrics.java +++ b/vespajlib/src/main/java/com/yahoo/concurrent/maintenance/JobMetrics.java @@ -1,7 +1,6 @@ // Copyright Verizon Media. Licensed under the terms of the Apache 2.0 license. See LICENSE in the project root. package com.yahoo.concurrent.maintenance; -import java.util.Map; import java.util.concurrent.ConcurrentHashMap; import java.util.function.BiConsumer; @@ -14,7 +13,7 @@ public class JobMetrics { private final BiConsumer<String, Long> metricConsumer; - private final Map<String, Long> incompleteRuns = new ConcurrentHashMap<>(); + private final ConcurrentHashMap<String, Long> incompleteRuns = new ConcurrentHashMap<>(); public JobMetrics(BiConsumer<String, Long> metricConsumer) { this.metricConsumer = metricConsumer; diff --git a/vespajlib/src/main/java/com/yahoo/concurrent/maintenance/Maintainer.java b/vespajlib/src/main/java/com/yahoo/concurrent/maintenance/Maintainer.java index 9fb5172ab0a..bd0e618ba05 100644 --- a/vespajlib/src/main/java/com/yahoo/concurrent/maintenance/Maintainer.java +++ b/vespajlib/src/main/java/com/yahoo/concurrent/maintenance/Maintainer.java @@ -27,18 +27,20 @@ public abstract class Maintainer implements Runnable { protected final Logger log = Logger.getLogger(this.getClass().getName()); private final String name; + private final Mode mode; private final JobControl jobControl; private final JobMetrics jobMetrics; private final Duration interval; private final ScheduledExecutorService service; private final AtomicBoolean shutDown = new AtomicBoolean(); - public Maintainer(String name, Duration interval, Instant startedAt, JobControl jobControl, JobMetrics jobMetrics, List<String> clusterHostnames) { - this(name, interval, staggeredDelay(interval, startedAt, HostName.getLocalhost(), clusterHostnames), jobControl, jobMetrics); + public Maintainer(String name, Mode mode, Duration interval, Instant startedAt, JobControl jobControl, JobMetrics jobMetrics, List<String> clusterHostnames) { + this(name, mode, interval, staggeredDelay(interval, startedAt, HostName.getLocalhost(), clusterHostnames), jobControl, jobMetrics); } - public Maintainer(String name, Duration interval, Duration initialDelay, JobControl jobControl, JobMetrics jobMetrics) { + public Maintainer(String name, Mode mode, Duration interval, Duration initialDelay, JobControl jobControl, JobMetrics jobMetrics) { this.name = name; + this.mode = Objects.requireNonNull(mode); this.interval = requireInterval(interval); this.jobControl = Objects.requireNonNull(jobControl); this.jobMetrics = Objects.requireNonNull(jobMetrics); @@ -49,13 +51,10 @@ public abstract class Maintainer implements Runnable { @Override public void run() { - log.log(Level.FINE, () -> "Running " + this.getClass().getSimpleName()); try { if (jobControl.isActive(name())) { lockAndMaintain(); } - } catch (UncheckedTimeoutException ignored) { - // Another actor is running this job } catch (Throwable e) { log.log(Level.WARNING, this + " failed. Will retry in " + interval.toMinutes() + " minutes", e); } @@ -94,14 +93,17 @@ public abstract class Maintainer implements Runnable { /** Run this while holding the job lock */ @SuppressWarnings("unused") public final void lockAndMaintain() { + log.log(Level.FINE, () -> "Running " + this.getClass().getSimpleName()); + jobMetrics.recordRunOf(name()); try (var lock = jobControl.lockJob(name())) { - try { - jobMetrics.recordRunOf(name()); - if (maintain()) jobMetrics.recordSuccessOf(name()); - } finally { - // Always forward metrics - jobMetrics.forward(name()); + if (maintain()) jobMetrics.recordSuccessOf(name()); + } catch (UncheckedTimeoutException ignored) { + if (mode == Mode.shared) { + // This is fine as we're colliding with a run on another node + jobMetrics.recordSuccessOf(name()); } + } finally { + jobMetrics.forward(name()); } } @@ -127,4 +129,14 @@ public abstract class Maintainer implements Runnable { return interval; } + public enum Mode { + + /** Completing a scheduled run on any node is sufficient */ + shared, + + /** Completing a scheduled run is always required */ + exclusive, + + } + } |