@@ -74,28 +74,50 @@ public class ThreadPool implements ReportingService<ThreadPoolInfo>, Scheduler,
7474 */
7575 public static class Names {
7676 /**
77- * All the tasks that do not relate to the purpose of one of the other thread pools should use this thread pool. Try to pick one of
78- * the other more specific thread pools where possible.
77+ * A thread pool with a very high (but finite) maximum size for when there really is no other choice.
78+ * <p>
79+ * This pool may be used for one-off CPU-bound activities, but the maximum size is so high that it doesn't really work well to do a
80+ * lot of CPU-bound work in parallel here. Likewise you can do IO on this pool, but using it for lots of concurrent IO is likely
81+ * harmful in clusters with poor concurrent IO performance (especially if using spinning disks). Blocking on a future on this pool
82+ * risks deadlock if there's a chance that the completion of the future depends on work being done on this pool. Unfortunately
83+ * that's pretty likely in most cases because of how often this pool is used; it's really rare because of the high limit on the pool
84+ * size, but when it happens it is extremely harmful to the node.
85+ * <p>
86+ * This pool is also used for recovery-related work. The recovery subsystem bounds its own concurrency, and therefore the amount of
87+ * recovery work done on the {@code #GENERIC} pool, via {@code cluster.routing.allocation.node_concurrent_recoveries} and related
88+ * settings.
89+ * <p>
90+ * This pool does not reject any task. If you submit a task after the pool starts to shut down, it may simply never run.
7991 */
8092 public static final String GENERIC = "generic" ;
93+
8194 /**
82- * Important management tasks that keep the cluster from falling apart.
83- * This thread pool ensures cluster coordination tasks do not get blocked by less critical tasks and can continue to make progress.
84- * This thread pool also defaults to a single thread, reducing contention on the Coordinator mutex.
95+ * A thread pool solely for the use of the cluster coordination subsystem that relates to cluster state updates, master elections,
96+ * cluster membership and so on.
97+ * <p>
98+ * This pool defaults to a single thread to avoid contention on {@code Coordinator#mutex}.
8599 */
86100 public static final String CLUSTER_COORDINATION = "cluster_coordination" ;
101+
87102 public static final String GET = "get" ;
88103 public static final String ANALYZE = "analyze" ;
89104 public static final String WRITE = "write" ;
90105 public static final String SEARCH = "search" ;
91106 public static final String SEARCH_COORDINATION = "search_coordination" ;
92107 public static final String AUTO_COMPLETE = "auto_complete" ;
93108 public static final String SEARCH_THROTTLED = "search_throttled" ;
109+
94110 /**
95- * Cluster management tasks. Tasks that manage data, and tasks that report on cluster health via statistics etc.
96- * Not a latency sensitive thread pool: some tasks may time be long-running; and the thread pool size is limited / relatively small.
111+ * A thread pool for running tasks related to cluster management, including collecting and exposing stats in APIs and certain other
112+ * internal tasks.
113+ * <p>
114+ * This pool is deliberately small in order to throttle the rate at which such tasks are executed and avoid diverting resources away
115+ * from production-critical work such as indexing and search. You may run long-running (CPU-bound or IO-bound) tasks on this pool,
116+ * but if the work relates to a REST API call then it must be cancellable in order to prevent an overexcited client from blocking or
117+ * delaying other management work.
97118 */
98119 public static final String MANAGEMENT = "management" ;
120+
99121 public static final String FLUSH = "flush" ;
100122 public static final String REFRESH = "refresh" ;
101123 public static final String WARMER = "warmer" ;
0 commit comments