docker stats --all: remove containers when removed

Before this patch, running `docker stats --all` would continue showing
all containers once observed. For example;

    CONTAINER ID   NAME                 CPU %     MEM USAGE / LIMIT     MEM %     NET I/O         BLOCK I/O        PIDS
    f2a785b0cd5f   foo1                 0.00%     8.535MiB / 7.653GiB   0.11%     1.12kB / 126B   0B / 12.3kB      11
    fc191b27517f   foo2                 0.00%     8.531MiB / 7.653GiB   0.11%     998B / 126B     0B / 12.3kB      11
    5040185fba53   foo3                 0.00%     8.578MiB / 7.653GiB   0.11%     872B / 126B     0B / 8.19kB      11

WHen removing `foo2`, the container would continue to be listed:

    CONTAINER ID   NAME                 CPU %     MEM USAGE / LIMIT     MEM %     NET I/O         BLOCK I/O        PIDS
    f2a785b0cd5f   foo1                 0.00%     8.535MiB / 7.653GiB   0.11%     1.12kB / 126B   0B / 12.3kB      11
    fc191b27517f   foo2                 --        -- / --               --        --              --               --
    5040185fba53   foo3                 0.00%     8.578MiB / 7.653GiB   0.11%     872B / 126B     0B / 12.3kB      11

Starting a new `foo2` container would now produce multiple entries:

    CONTAINER ID   NAME                 CPU %     MEM USAGE / LIMIT     MEM %     NET I/O         BLOCK I/O        PIDS
    f2a785b0cd5f   foo1                 0.00%     8.535MiB / 7.653GiB   0.11%     1.25kB / 126B   0B / 12.3kB      11
    fc191b27517f   foo2                 --        -- / --               --        --              --               --
    5040185fba53   foo3                 0.00%     8.578MiB / 7.653GiB   0.11%     998B / 126B     0B / 12.3kB      11
    dba11b9e1ba9   foo2                 0.00%     8.578MiB / 7.653GiB   0.11%     872B / 126B     0B / 8.19kB      11

Repeat that, and the list would continue to grow;

    CONTAINER ID   NAME                 CPU %     MEM USAGE / LIMIT     MEM %     NET I/O         BLOCK I/O        PIDS
    f2a785b0cd5f   foo1                 0.00%     8.535MiB / 7.653GiB   0.11%     1.25kB / 126B   0B / 12.3kB      11
    fc191b27517f   foo2                 --        -- / --               --        --              --               --
    5040185fba53   foo3                 0.00%     8.578MiB / 7.653GiB   0.11%     998B / 126B     0B / 12.3kB      11
    dba11b9e1ba9   foo2                 --        -- / --               --        --              --               --
    193a6dcfaa2d   foo2                 --        -- / --               --        --              --               --
    bf50e58085c6   foo2                 0.00%     8.539MiB / 7.653GiB   0.11%     872B / 126B     0B / 8.19kB      11

After this patch, containers are removed when we observe a `destroy` event;

    CONTAINER ID   NAME                 CPU %     MEM USAGE / LIMIT     MEM %     NET I/O         BLOCK I/O        PIDS
    f2a785b0cd5f   foo1                 0.00%     8.535MiB / 7.653GiB   0.11%     1.5kB / 126B    0B / 12.3kB      11
    5040185fba53   foo3                 0.00%     8.578MiB / 7.653GiB   0.11%     1.25kB / 126B   0B / 12.3kB      11
    bf50e58085c6   foo2                 0.00%     8.539MiB / 7.653GiB   0.11%     872B / 126B     0B / 12.3kB      11

Containers are added when created, so in the example above, the new `foo2`
is added at the end:

    CONTAINER ID   NAME                 CPU %     MEM USAGE / LIMIT     MEM %     NET I/O         BLOCK I/O        PIDS
    f2a785b0cd5f   foo1                 0.00%     8.535MiB / 7.653GiB   0.11%     1.5kB / 126B    0B / 12.3kB      11
    5040185fba53   foo3                 0.00%     8.578MiB / 7.653GiB   0.11%     1.25kB / 126B   0B / 12.3kB      11
    bf50e58085c6   foo2                 0.00%     8.539MiB / 7.653GiB   0.11%     872B / 126B     0B / 12.3kB      11

If a container dies, and `--all` is set, we continue listing it, but stats
are not updated while the container is stopped (we should consider resetting
the stats and show `-- / --` to be more clear that we don't have the container
running).

Here's with `foo3` stopped:

    CONTAINER ID   NAME                 CPU %     MEM USAGE / LIMIT     MEM %     NET I/O         BLOCK I/O        PIDS
    f2a785b0cd5f   foo1                 0.00%     8.535MiB / 7.653GiB   0.11%     1.5kB / 126B    0B / 12.3kB      11
    5040185fba53   foo3                 0.00%     0B / 0B               0.00%     0B / 0B         0B / 0B          0
    bf50e58085c6   foo2                 0.00%     8.539MiB / 7.653GiB   0.11%     872B / 126B     0B / 12.3kB      11

Starting the container continues updating its stats:

    CONTAINER ID   NAME                 CPU %     MEM USAGE / LIMIT     MEM %     NET I/O         BLOCK I/O        PIDS
    f2a785b0cd5f   foo1                 0.00%     8.535MiB / 7.653GiB   0.11%     1.63kB / 126B   0B / 12.3kB      11
    5040185fba53   foo3                 0.00%     8.496MiB / 7.653GiB   0.11%     872B / 126B     0B / 0B          11
    bf50e58085c6   foo2                 0.00%     8.539MiB / 7.653GiB   0.11%     998B / 126B     0B / 12.3kB      11

When running without `--all`, we continue to remove containers as soon as
possible (`die` events), but with `--all`, those events are ignored with
the expectation that the container might come back.

Signed-off-by: Sebastiaan van Stijn <github@gone.nl>

Author: thaJeztah

cli/command/container/stats.go

@@ -156,12 +156,20 @@ func RunStats(ctx context.Context, dockerCLI command.Cli, options *StatsOptions)
 			}
 		})
 
+		// Remove containers when they are removed ("destroyed"); containers
+		// do not emit [events.ActionRemove], only [events.ActionDestroy].
+		//
+		// When running with "--all" we don't remove containers when they die,
+		// because they may come back, but without "--all" we remove them
+		// on the first possible occasion (either "die" or "destroy").
+		rmEvents := []events.Action{events.ActionDestroy}
 		if !options.All {
-			eh.setHandler([]events.Action{events.ActionDie}, func(ctx context.Context, e events.Message) {
-				log.G(ctx).Debug("stop collecting stats for container")
-				cStats.remove(e.Actor.ID)
-			})
+			rmEvents = append(rmEvents, events.ActionDie)
 		}
+		eh.setHandler(rmEvents, func(ctx context.Context, e events.Message) {
+			log.G(ctx).Debug("stop collecting stats for container")
+			cStats.remove(e.Actor.ID)
+		})
 
 		// monitorContainerEvents watches for container creation and removal (only
 		// used when calling `docker stats` without arguments).