Skip to contentSkip to conversation

OCFS2: allow the ocfs2 heartbeat thread to prioritize I/O --revision 3

PATCH New Zhen Wei 2 participants 2 messages Dec 7, 2006 – Dec 8, 2006 OCFS2

lore
Share & tools
lore

Zhen Wei

author
#0 lore raw

PATCH 2/3 OCFS2: allow the ocfs2 heartbeat thread to prioritize I/O --revision 3

- replace calling syscall interface with running ionice
- fix the running ionice return code
- use device name instead of uuid for mount.ocfs2

zhen wei
[email protected]
+86 10 65339225

-------------- next part --------------

libocfs2/ocfs2_err.et +3 -0
@@ -159,4 +159,7 @@ ec	OCFS2_ET_BLOCK_SIZE_TOO_SMALL_FOR_HAR
 ec	OCFS2_ET_INVALID_LOCKRES,
 	"The lock name is invalid"
 
+ec	OCFS2_ET_NO_IONICE,
+	"Can't find ionice"
+
 	end
libo2cb/include/o2cb.h +4 -0
@@ -98,6 +98,10 @@ errcode_t o2cb_start_heartbeat_region_pe
 errcode_t o2cb_stop_heartbeat_region_perm(const char *cluster_name,
 					  const char *region_name);
 
+errcode_t o2cb_get_hb_thread_pid (const char *cluster_name, 
+				  const char *region_name, 
+				  pid_t *pid);
+
 errcode_t o2cb_get_region_ref(const char *region_name,
 			      int undo);
 errcode_t o2cb_put_region_ref(const char *region_name,
libo2cb/o2cb_abi.c +30 -0
@@ -1257,6 +1257,36 @@ void o2cb_free_nodes_list(char **nodes)
 	o2cb_free_dir_list(nodes);
 }
 
+errcode_t o2cb_get_hb_thread_pid (const char *cluster_name, const char *region_name,
+			   pid_t *pid)
+{
+	char attr_path[PATH_MAX];
+	char _fake_cluster_name[NAME_MAX];
+	char attr_value[16];
+	errcode_t ret;
+
+	if (!cluster_name) {
+		ret = _fake_default_cluster(_fake_cluster_name);
+		if (ret)
+			return ret;
+		cluster_name = _fake_cluster_name;
+	}
+
+	ret = snprintf(attr_path, PATH_MAX - 1,
+		       O2CB_FORMAT_HEARTBEAT_REGION_ATTR,
+		       configfs_path, cluster_name, region_name,
+		       "pid");
+
+	if ((ret <= 0) || (ret == (PATH_MAX - 1)))
+		return O2CB_ET_INTERNAL_FAILURE;
+
+	ret = o2cb_get_attribute(attr_path, attr_value, sizeof(attr_value) - 1);
+	if (ret == 0)
+		*pid = atoi (attr_value);
+
+	return ret;
+}
+
 errcode_t o2cb_get_node_num(const char *cluster_name, const char *node_name,
 			    uint16_t *node_num)
 {
ocfs2_hb_ctl/ocfs2_hb_ctl.c +65 -2

133 lines not on this page · raw message

#1 lore raw

PATCH 2/3 OCFS2: allow the ocfs2 heartbeat thread to prioritize I/O --revision 3

Hi Zwei,

Quoted 3 lines · Zhen Wei

- replace calling syscall interface with running ionice
- fix the running ionice return code
- use device name instead of uuid for mount.ocfs2

The kernel patch and the mount.ocfs2 patches look good. I only have a couple of comments for this one below.

ocfs2_hb_ctl/ocfs2_hb_ctl.c static errcode_t start_heartbeat(struct · line 301

Quoted 11 lines
ocfs2_hb_ctl/ocfs2_hb_ctl.c +7 -0
@@ -289,6 +292,41 @@ static errcode_t start_heartbeat(struct 
 	return err;
 }
 
+static errcode_t adjust_priority(struct hb_ctl_options *hbo)
+{
+	int ret, child_status;
+	pid_t hb_pid, child_pid;
+	char level_arg[16], pid_arg[16];
+
+	if (access ("/usr/bin/ionice", X_OK) != 0)

You might want to make "/usr/bin/ionice" a #define at the top of this file, since we use it at least twice.

ocfs2_hb_ctl/ocfs2_hb_ctl.c static errcode_t start_heartbeat(struct · line 304

Quoted 3 lines
ocfs2_hb_ctl/ocfs2_hb_ctl.c +3 -0
+		return OCFS2_ET_NO_IONICE;
+
+	get_uuid (hbo->dev_str, hbo->uuid_str);

You shouldn't have to call get_uuid() here -- see my comments below.

ocfs2_hb_ctl/ocfs2_hb_ctl.c static int process_options(struct hb_ctl · line 440

Quoted 10 lines
ocfs2_hb_ctl/ocfs2_hb_ctl.c +6 -0
@@ -385,6 +432,13 @@ static int process_options(struct hb_ctl
 			ret = -EINVAL;
 		break;
 
+	case HB_ACTION_IONICE:
+		/* ionice needs uuid and priority */
+		if (!hbo->dev_str || hbo->io_prio < 0 ||
+		    hbo->io_prio > 7)
+			ret = -EINVAL;
+		break;

The ionice option should be just like the other options and allow the user to specify an uuid _or_ a device. The main function will call get_uuid() if the device was specified.

This will also help avoid the manual call to get_uuid() you have up in adjust_priority().

ocfs2_hb_ctl/ocfs2_hb_ctl.c int main(int argc, char **argv) · line 546

Quoted 10 lines
ocfs2_hb_ctl/ocfs2_hb_ctl.c +6 -0
@@ -483,6 +538,14 @@ int main(int argc, char **argv)
 		}
 		break;
 
+	case HB_ACTION_IONICE:
+		err = adjust_priority(&hbo);
+		if (err) {
+			com_err(progname, err, "while adjusting heartbeat thread I/O priority");
+			ret = err;
+		}

You're still printing errors unconditionally.

To break it down, the two errors we don't want to print for are:

* Not finding that the heartbeat region 'pid' attribute exists in configfs. (Should be an O2CB_ET_SERVICE_UNAVAILABLE return)

* Not being able to run /usr/bin/ionice because it doesn't exist. (Should be an OCFS2_ET_NO_IONICE return)

Neither of those two errors should be fatal, or should print to the user. You'll have to filter them from the com_err.

Right now the rule is that all versions of ocfs2-tools must be compatible with old versions of the file system module.

Thanks for the patches so far,

	--Mark

--
Mark Fasheh
Senior Software Developer, Oracle
[email protected]