public inbox for pve-devel@lists.proxmox.com
 help / color / mirror / Atom feed
* [pve-devel] [PATCH qemu-server v4 0/2] qmeventd: improve shutdown behaviour
@ 2022-09-23  9:51 Dominik Csapak
  2022-09-23  9:51 ` [pve-devel] [PATCH qemu-server v4 1/2] qmeventd: rework 'forced_cleanup' handling and set timeout to 60s Dominik Csapak
                   ` (2 more replies)
  0 siblings, 3 replies; 4+ messages in thread
From: Dominik Csapak @ 2022-09-23  9:51 UTC (permalink / raw)
  To: pve-devel

includes the following improvements:
* increases 'force cleanup' timeout to 60s (from 5)
* saves individual timeout for each vm
* don't force cleanup for vms where normal cleanup worked
* sending QMP quit instead of SIGTERM (less log noise)

changes from v3:
* merge CleanupData into Client (pidfd,timeout) and use that for the
   forced_cleanup list
* close pidfd also in cleanup_client (and remove itself from the
   forced_cleanup list)
* guard the sigkill with a check for timeout != 0
* with those changes patch 2/3 is not necessary anymore

changes from v2:
* change from cast of the function to cast of the parameters
* don't copy the variable on the stack in 'sigkill'

changes from v1:
* remove 'alarm' calls altogether and use epoll_waits' timeout mechanic
   instead
* call 'time()' only once and give it as user data to the function
* change the function singatures and cast on callsite with '(GFunc)'
   for the g_slist_foreach calls
* change to <s> for the usage output for timeouts

Dominik Csapak (2):
  qmeventd: rework 'forced_cleanup' handling and set timeout to 60s
  qmeventd: send QMP 'quit' command instead of SIGTERM

 qmeventd/qmeventd.c | 112 ++++++++++++++++++++++++--------------------
 qmeventd/qmeventd.h |   8 ++--
 2 files changed, 65 insertions(+), 55 deletions(-)

-- 
2.30.2





^ permalink raw reply	[flat|nested] 4+ messages in thread

* [pve-devel] [PATCH qemu-server v4 1/2] qmeventd: rework 'forced_cleanup' handling and set timeout to 60s
  2022-09-23  9:51 [pve-devel] [PATCH qemu-server v4 0/2] qmeventd: improve shutdown behaviour Dominik Csapak
@ 2022-09-23  9:51 ` Dominik Csapak
  2022-09-23  9:51 ` [pve-devel] [PATCH qemu-server v4 2/2] qmeventd: send QMP 'quit' command instead of SIGTERM Dominik Csapak
  2022-09-23 10:20 ` [pve-devel] applied-series: [PATCH qemu-server v4 0/2] qmeventd: improve shutdown behaviour Wolfgang Bumiller
  2 siblings, 0 replies; 4+ messages in thread
From: Dominik Csapak @ 2022-09-23  9:51 UTC (permalink / raw)
  To: pve-devel

currently, the 'forced_cleanup' (sending SIGKILL to the qemu process),
is intended to be triggered 5 seconds after sending the initial shutdown
signal (SIGTERM) which is sadly not enough for some setups.

Accidentally, it could be triggered earlier than 5 seconds, if a
SIGALRM triggers in the timespan directly before setting it again.

Also, this approach means that depending on when machines are shutdown
their forced cleanup may happen after 5 seconds, or any time after, if
new vms are shut off in the meantime.

Improve this situation by reworking the way we deal with this cleanup.
We save the pidfd, time incl. timeout in the Client, and set a timeout
to 'epoll_wait' of 10 seconds, which will then trigger a forced_cleanup.
Remove entries from the forced_cleanup list when that entry is killed,
or when the normal cleanup took place.

To improve the shutdown behaviour, increase the default timeout to 60
seconds, which should be enough, but add a commandline toggle where
users can set it to a different value.

Signed-off-by: Dominik Csapak <d.csapak@proxmox.com>
---
 qmeventd/qmeventd.c | 98 +++++++++++++++++++++++----------------------
 qmeventd/qmeventd.h |  8 ++--
 2 files changed, 54 insertions(+), 52 deletions(-)

diff --git a/qmeventd/qmeventd.c b/qmeventd/qmeventd.c
index 8d32827..3256d42 100644
--- a/qmeventd/qmeventd.c
+++ b/qmeventd/qmeventd.c
@@ -29,6 +29,7 @@
 #include <signal.h>
 #include <stdbool.h>
 #include <stdio.h>
+#include <stdlib.h>
 #include <string.h>
 #include <sys/epoll.h>
 #include <sys/socket.h>
@@ -36,15 +37,19 @@
 #include <sys/un.h>
 #include <sys/wait.h>
 #include <unistd.h>
+#include <time.h>
 
 #include "qmeventd.h"
 
+#define DEFAULT_KILL_TIMEOUT 60
+
 static int verbose = 0;
+static int kill_timeout = DEFAULT_KILL_TIMEOUT;
 static int epoll_fd = 0;
 static const char *progname;
 GHashTable *vm_clients; // key=vmid (freed on remove), value=*Client (free manually)
 GSList *forced_cleanups;
-volatile sig_atomic_t alarm_triggered = 0;
+static int needs_cleanup = 0;
 
 /*
  * Helper functions
@@ -56,6 +61,7 @@ usage()
     fprintf(stderr, "Usage: %s [-f] [-v] PATH\n", progname);
     fprintf(stderr, "  -f       run in foreground (default: false)\n");
     fprintf(stderr, "  -v       verbose (default: false)\n");
+    fprintf(stderr, "  -t <s>   kill timeout (default: %ds)\n", DEFAULT_KILL_TIMEOUT);
     fprintf(stderr, "  PATH     use PATH for socket\n");
 }
 
@@ -436,6 +442,11 @@ cleanup_client(struct Client *client)
 	    break;
     }
 
+    if (client->pidfd > 0) {
+	(void)close(client->pidfd);
+    }
+    VERBOSE_PRINT("removing %s from forced cleanups\n", client->qemu.vmid);
+    forced_cleanups = g_slist_remove(forced_cleanups, client);
     free(client);
 }
 
@@ -469,16 +480,13 @@ terminate_client(struct Client *client)
     int err = kill(client->pid, SIGTERM);
     log_neg(err, "kill");
 
-    struct CleanupData *data_ptr = malloc(sizeof(struct CleanupData));
-    struct CleanupData data = {
-	.pid = client->pid,
-	.pidfd = pidfd
-    };
-    *data_ptr = data;
-    forced_cleanups = g_slist_prepend(forced_cleanups, (void *)data_ptr);
+    time_t timeout = time(NULL) + kill_timeout;
+
+    client->pidfd = pidfd;
+    client->timeout = timeout;
 
-    // resets any other alarms, but will fire eventually and cleanup all
-    alarm(5);
+    forced_cleanups = g_slist_prepend(forced_cleanups, (void *)client);
+    needs_cleanup = 1;
 }
 
 void
@@ -551,58 +559,51 @@ handle_client(struct Client *client)
     json_tokener_free(tok);
 }
 
-
-/*
- * SIGALRM and cleanup handling
- *
- * terminate_client will set an alarm for 5 seconds and add its client's PID to
- * the forced_cleanups list - when the timer expires, we iterate the list and
- * attempt to issue SIGKILL to all processes which haven't yet stopped.
- */
-
-static void
-alarm_handler(__attribute__((unused)) int signum)
-{
-    alarm_triggered = 1;
-}
-
 static void
-sigkill(void *ptr, __attribute__((unused)) void *unused)
+sigkill(void *ptr, void *time_ptr)
 {
-    struct CleanupData data = *((struct CleanupData *)ptr);
+    struct Client *data = ptr;
     int err;
 
-    if (data.pidfd > 0) {
-	err = pidfd_send_signal(data.pidfd, SIGKILL, NULL, 0);
-	(void)close(data.pidfd);
+    if (data->timeout != 0 && data->timeout > *(time_t *)time_ptr) {
+	return;
+    }
+
+    if (data->pidfd > 0) {
+	err = pidfd_send_signal(data->pidfd, SIGKILL, NULL, 0);
+	(void)close(data->pidfd);
+	data->pidfd = -1;
     } else {
-	err = kill(data.pid, SIGKILL);
+	err = kill(data->pid, SIGKILL);
     }
 
     if (err < 0) {
 	if (errno != ESRCH) {
 	    fprintf(stderr, "SIGKILL cleanup of pid '%d' failed - %s\n",
-		    data.pid, strerror(errno));
+		    data->pid, strerror(errno));
 	}
     } else {
 	fprintf(stderr, "cleanup failed, terminating pid '%d' with SIGKILL\n",
-		data.pid);
+		data->pid);
     }
+
+    data->timeout = 0;
+
+    // remove ourselves from the list
+    forced_cleanups = g_slist_remove(forced_cleanups, ptr);
 }
 
 static void
 handle_forced_cleanup()
 {
-    if (alarm_triggered) {
+    if (g_slist_length(forced_cleanups) > 0) {
 	VERBOSE_PRINT("clearing forced cleanup backlog\n");
-	alarm_triggered = 0;
-	g_slist_foreach(forced_cleanups, sigkill, NULL);
-	g_slist_free_full(forced_cleanups, free);
-	forced_cleanups = NULL;
+	time_t cur_time = time(NULL);
+	g_slist_foreach(forced_cleanups, sigkill, &cur_time);
     }
+    needs_cleanup = g_slist_length(forced_cleanups) > 0;
 }
 
-
 int
 main(int argc, char *argv[])
 {
@@ -611,7 +612,7 @@ main(int argc, char *argv[])
     char *socket_path = NULL;
     progname = argv[0];
 
-    while ((opt = getopt(argc, argv, "hfv")) != -1) {
+    while ((opt = getopt(argc, argv, "hfvt:")) != -1) {
 	switch (opt) {
 	    case 'f':
 		daemonize = 0;
@@ -619,6 +620,15 @@ main(int argc, char *argv[])
 	    case 'v':
 		verbose = 1;
 		break;
+	    case 't':
+		errno = 0;
+		char *endptr = NULL;
+		kill_timeout = strtoul(optarg, &endptr, 10);
+		if (errno != 0 || *endptr != '\0' || kill_timeout == 0) {
+		    usage();
+		    exit(EXIT_FAILURE);
+		}
+		break;
 	    case 'h':
 		usage();
 		exit(EXIT_SUCCESS);
@@ -635,7 +645,6 @@ main(int argc, char *argv[])
     }
 
     signal(SIGCHLD, SIG_IGN);
-    signal(SIGALRM, alarm_handler);
 
     socket_path = argv[optind];
 
@@ -669,11 +678,7 @@ main(int argc, char *argv[])
     int nevents;
 
     for(;;) {
-	nevents = epoll_wait(epoll_fd, events, 1, -1);
-	if (nevents < 0 && errno == EINTR) {
-	    handle_forced_cleanup();
-	    continue;
-	}
+	nevents = epoll_wait(epoll_fd, events, 1, needs_cleanup ? 10*1000 : -1);
 	bail_neg(nevents, "epoll_wait");
 
 	for (int n = 0; n < nevents; n++) {
@@ -688,7 +693,6 @@ main(int argc, char *argv[])
 		handle_client((struct Client *)events[n].data.ptr);
 	    }
 	}
-
 	handle_forced_cleanup();
     }
 }
diff --git a/qmeventd/qmeventd.h b/qmeventd/qmeventd.h
index 2cf1947..9afc935 100644
--- a/qmeventd/qmeventd.h
+++ b/qmeventd/qmeventd.h
@@ -7,6 +7,7 @@
 */
 
 #include <sys/syscall.h>
+#include <time.h>
 
 #ifndef __NR_pidfd_open
 #define __NR_pidfd_open 434
@@ -63,6 +64,8 @@ struct Client {
 
     int fd;
     pid_t pid;
+    int pidfd;
+    time_t timeout;
 
     ClientType type;
     ClientState state;
@@ -83,11 +86,6 @@ struct Client {
     } vzdump;
 };
 
-struct CleanupData {
-    pid_t pid;
-    int pidfd;
-};
-
 void handle_qmp_handshake(struct Client *client);
 void handle_qmp_event(struct Client *client, struct json_object *obj);
 void handle_qmp_return(struct Client *client, struct json_object *data, bool error);
-- 
2.30.2





^ permalink raw reply	[flat|nested] 4+ messages in thread

* [pve-devel] [PATCH qemu-server v4 2/2] qmeventd: send QMP 'quit' command instead of SIGTERM
  2022-09-23  9:51 [pve-devel] [PATCH qemu-server v4 0/2] qmeventd: improve shutdown behaviour Dominik Csapak
  2022-09-23  9:51 ` [pve-devel] [PATCH qemu-server v4 1/2] qmeventd: rework 'forced_cleanup' handling and set timeout to 60s Dominik Csapak
@ 2022-09-23  9:51 ` Dominik Csapak
  2022-09-23 10:20 ` [pve-devel] applied-series: [PATCH qemu-server v4 0/2] qmeventd: improve shutdown behaviour Wolfgang Bumiller
  2 siblings, 0 replies; 4+ messages in thread
From: Dominik Csapak @ 2022-09-23  9:51 UTC (permalink / raw)
  To: pve-devel

this is functionally the same, but sending SIGTERM has the ugly side
effect of printing the following to the log:

> QEMU[<pid>]: kvm: terminating on signal 15 from pid <pid> (/usr/sbin/qmeventd)

while sending a QMP quit command does not.

Signed-off-by: Dominik Csapak <d.csapak@proxmox.com>
---
 qmeventd/qmeventd.c | 14 +++++++++++---
 1 file changed, 11 insertions(+), 3 deletions(-)

diff --git a/qmeventd/qmeventd.c b/qmeventd/qmeventd.c
index 3256d42..499376a 100644
--- a/qmeventd/qmeventd.c
+++ b/qmeventd/qmeventd.c
@@ -287,8 +287,10 @@ handle_qmp_return(struct Client *client, struct json_object *data, bool error)
 	    VERBOSE_PRINT("%s: QMP handshake complete\n", client->qemu.vmid);
 	    break;
 
-	case STATE_IDLE:
+	// we expect an empty return object after sending quit
 	case STATE_TERMINATING:
+	    break;
+	case STATE_IDLE:
 	    VERBOSE_PRINT("%s: spurious return value received\n",
 			  client->qemu.vmid);
 	    break;
@@ -477,8 +479,14 @@ terminate_client(struct Client *client)
 	}
     }
 
-    int err = kill(client->pid, SIGTERM);
-    log_neg(err, "kill");
+    // try to send a 'quit' command first, fallback to SIGTERM of the pid
+    static const char qmp_quit_command[] = "{\"execute\":\"quit\"}\n";
+    VERBOSE_PRINT("%s: sending 'quit' via QMP\n", client->qemu.vmid);
+    if (!must_write(client->fd, qmp_quit_command, sizeof(qmp_quit_command) - 1)) {
+	VERBOSE_PRINT("%s: sending 'SIGTERM' to pid %d\n", client->qemu.vmid, client->pid);
+	int err = kill(client->pid, SIGTERM);
+	log_neg(err, "kill");
+    }
 
     time_t timeout = time(NULL) + kill_timeout;
 
-- 
2.30.2





^ permalink raw reply	[flat|nested] 4+ messages in thread

* [pve-devel] applied-series: [PATCH qemu-server v4 0/2] qmeventd: improve shutdown behaviour
  2022-09-23  9:51 [pve-devel] [PATCH qemu-server v4 0/2] qmeventd: improve shutdown behaviour Dominik Csapak
  2022-09-23  9:51 ` [pve-devel] [PATCH qemu-server v4 1/2] qmeventd: rework 'forced_cleanup' handling and set timeout to 60s Dominik Csapak
  2022-09-23  9:51 ` [pve-devel] [PATCH qemu-server v4 2/2] qmeventd: send QMP 'quit' command instead of SIGTERM Dominik Csapak
@ 2022-09-23 10:20 ` Wolfgang Bumiller
  2 siblings, 0 replies; 4+ messages in thread
From: Wolfgang Bumiller @ 2022-09-23 10:20 UTC (permalink / raw)
  To: Dominik Csapak; +Cc: pve-devel

applied both patches, thanks

On Fri, Sep 23, 2022 at 11:51:13AM +0200, Dominik Csapak wrote:
> includes the following improvements:
> * increases 'force cleanup' timeout to 60s (from 5)
> * saves individual timeout for each vm
> * don't force cleanup for vms where normal cleanup worked
> * sending QMP quit instead of SIGTERM (less log noise)
> 
> changes from v3:
> * merge CleanupData into Client (pidfd,timeout) and use that for the
>    forced_cleanup list
> * close pidfd also in cleanup_client (and remove itself from the
>    forced_cleanup list)
> * guard the sigkill with a check for timeout != 0
> * with those changes patch 2/3 is not necessary anymore
> 
> changes from v2:
> * change from cast of the function to cast of the parameters
> * don't copy the variable on the stack in 'sigkill'
> 
> changes from v1:
> * remove 'alarm' calls altogether and use epoll_waits' timeout mechanic
>    instead
> * call 'time()' only once and give it as user data to the function
> * change the function singatures and cast on callsite with '(GFunc)'
>    for the g_slist_foreach calls
> * change to <s> for the usage output for timeouts
> 
> Dominik Csapak (2):
>   qmeventd: rework 'forced_cleanup' handling and set timeout to 60s
>   qmeventd: send QMP 'quit' command instead of SIGTERM
> 
>  qmeventd/qmeventd.c | 112 ++++++++++++++++++++++++--------------------
>  qmeventd/qmeventd.h |   8 ++--
>  2 files changed, 65 insertions(+), 55 deletions(-)
> 
> -- 
> 2.30.2




^ permalink raw reply	[flat|nested] 4+ messages in thread

end of thread, other threads:[~2022-09-23 10:21 UTC | newest]

Thread overview: 4+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2022-09-23  9:51 [pve-devel] [PATCH qemu-server v4 0/2] qmeventd: improve shutdown behaviour Dominik Csapak
2022-09-23  9:51 ` [pve-devel] [PATCH qemu-server v4 1/2] qmeventd: rework 'forced_cleanup' handling and set timeout to 60s Dominik Csapak
2022-09-23  9:51 ` [pve-devel] [PATCH qemu-server v4 2/2] qmeventd: send QMP 'quit' command instead of SIGTERM Dominik Csapak
2022-09-23 10:20 ` [pve-devel] applied-series: [PATCH qemu-server v4 0/2] qmeventd: improve shutdown behaviour Wolfgang Bumiller

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox
Service provided by Proxmox Server Solutions GmbH | Privacy | Legal