src/host_id.c src/token_manager.c

[David Teigland]

 src/host_id.c       |    5 ++++-
 src/token_manager.c |   28 ++++++++++++++++++++++++----
 2 files changed, 28 insertions(+), 5 deletions(-)

New commits:
commit 3d14379e4aea8eef07cfd0ff7dee586a89ef4b9d
Author: David Teigland <teigland(a)redhat.com&gt;
Date:   Tue Sep 13 14:40:34 2011 -0500

    sanlock: SIGTERM and SIGKILL for REQ_KILL_PID
    
    and add some comments about another req mode

diff --git a/src/host_id.c b/src/host_id.c
index 4d39d82..18ef28d 100644
--- a/src/host_id.c
+++ b/src/host_id.c
@@ -519,6 +519,7 @@ static void *lockspace_thread(void *arg_in)
 
 		/*
 		 * pet the watchdog
+		 * (don't update on thread_stop because it's probably unlinked)
 		 */
 
 		if (delta_result == SANLK_OK && !sp->thread_stop)
@@ -543,7 +544,9 @@ static void *lockspace_thread(void *arg_in)
 		}
 	}
 
-	/* unlink called below to get it done ASAP */
+	/* watchdog unlink was done in main_loop when thread_stop was set, to
+	   get it done as quickly as possible in case the wd is about to fire. */
+
 	close_watchdog_file(sp);
  out:
 	if (delta_result == SANLK_OK)
diff --git a/src/token_manager.c b/src/token_manager.c
index 6568af3..33de66a 100644
--- a/src/token_manager.c
+++ b/src/token_manager.c
@@ -440,15 +440,35 @@ static void *resource_thread(void *arg GNUC_UNUSED)
 				continue;
 			}
 
+			/*
+			 * TODO: add force_mode SANLK_REQ_KILL_PID_OR_RESET
+			 * which would attempt to kill the pid like KILL_PID,
+			 * but if the pid doesn't exit will block watchdog
+			 * updates to reset the host.
+			 * Here set r->block_wd_time = now + pid_exit_time,
+			 * In renewal check for any r in resources with
+			 * block_wd_time <= now, and if found will not
+			 * update the watchdog.  If the pid continues to
+			 * not exit, the wd will fire and reset the machine.
+			 * If the pid exits before pid_exit_time, no wd
+			 * updates will be skipped.
+			 */
+
 			if (req.force_mode == SANLK_REQ_KILL_PID) {
-				/* look up r again to check it still exists and
-				   pid is same? */
+				/* look up r again? verify it exists and pid same */
 
-				log_error("req_kill_pid %d %.48s:%.48s", pid,
+				log_debug("req force_mode %u pid %d %.48s:%.48s",
+					  req.force_mode, pid,
 					  tt->r.lockspace_name, tt->r.name);
+
+				/* TODO: share code with kill_pids() to gradually
+				 * escalate from killscript, SIGTERM, SIGKILL */
+
+				kill(pid, SIGTERM);
+				sleep(1);
 				kill(pid, SIGKILL);
 			} else {
-				log_error("req force_mode unknown %u", req.force_mode);
+				log_error("req force_mode %u unknown", req.force_mode);
 			}
 		}
 	}