Skip to contentSkip to conversation

multipath.conf(5): remove io-affinity information

PATCH In review Benjamin Marzinski 4 participants 6 messages Nov 30, 2022 – Dec 7, 2022 Device mapper

lore
Share & tools
lore

Benjamin Marzinski

author maintainer
#0 lore raw

PATCH 1/2 multipath.conf(5): remove io-affinity information

The multpath-tools do not support the io-affinity path selector. We always add a repeat count as the path argument. The io-affinity selector doesn't take one. Instead it takes a bitmap of CPUs that a path can run on. This isn't something that lends itself to the kind of auto-assembling that multipathd does. But even if we did want to try to support this path-selector, until we do, we shouldn't be listing it in the multipath.conf documentation.

---
 multipath/multipath.conf.5 | 4 ----
 1 file changed, 4 deletions(-)
multipath/multipath.conf.5 +0 -4
index 1fea9d5a..3a45ac89 100644
@@ -205,10 +205,6 @@ of outstanding I/O to the path and its relative throughput.
 estimation of future service time based on the history of previous I/O submitted
 to each path.
 .TP
-.I "io-affinity 0"
-(Since 5.11 kernel) Choose the path for the next bunch of I/O based on a CPU to
-path mapping the user passes in and what CPU we are executing on.
-.TP
 The default is: \fBservice-time 0\fR
 .RE
 .
Signature · 6 lines
2.17.2

--
dm-devel mailing list
[email protected]
https://listman.redhat.com/mailman/listinfo/dm-devel
author maintainer
#1 lore raw

PATCH 2/2 libmpathpersist: fix command keyword ordering

When libmpathpersist was communicating with multipathd, it wasn't using the correct keyword order for the commands, as specified in the CLI commands reference. Since commit f812466f, multipathd requires commands to be ordered correctly. Fix the ordering.

Fixes f812466f ("multipathd: more robust command parsing")
Cc lixiaokeng <[email protected]>
---
 libmpathpersist/mpath_updatepr.c | 23 ++++++++++++-----------
 1 file changed, 12 insertions(+), 11 deletions(-)
libmpathpersist/mpath_updatepr.c +12 -11
index 5824c169..4529a82b 100644
@@ -18,7 +18,7 @@
 #include "mpathpr.h"
 
 
-static int do_update_pr(char *alias, char *arg)
+static int do_update_pr(char *alias, char *cmd, char *key)
 {
 	int fd;
 	char str[256];
@@ -31,7 +31,10 @@ static int do_update_pr(char *alias, char *arg)
 		return -1;
 	}
 
-	snprintf(str,sizeof(str),"map %s %s", alias, arg);
+	if (key)
+		snprintf(str,sizeof(str),"%s map %s key %s", cmd, alias, key);
+	else
+		snprintf(str,sizeof(str),"%s map %s", cmd, alias);
 	condlog (2, "%s: pr message=%s", alias, str);
 	if (send_packet(fd, str) != 0) {
 		condlog(2, "%s: message=%s send error=%d", alias, str, errno);
@@ -56,18 +59,16 @@ static int do_update_pr(char *alias, char *arg)
 }
 
 int update_prflag(char *mapname, int set) {
-	return do_update_pr(mapname, (set)? "setprstatus" : "unsetprstatus");
+	return do_update_pr(mapname, (set)? "setprstatus" : "unsetprstatus",
+			    NULL);
 }
 
 int update_prkey_flags(char *mapname, uint64_t prkey, uint8_t sa_flags) {
 	char str[256];
-	char *flagstr = "";
 
-	if (sa_flags & MPATH_F_APTPL_MASK)
-		flagstr = ":aptpl";
-	if (prkey)
-		sprintf(str, "setprkey key %" PRIx64 "%s", prkey, flagstr);
-	else
-		sprintf(str, "unsetprkey");
-	return do_update_pr(mapname, str);
+	if (!prkey)
+		return do_update_pr(mapname, "unsetprkey", NULL);
+	sprintf(str, "%" PRIx64 "%s", prkey,
+		(sa_flags & MPATH_F_APTPL_MASK) ? ":aptpl" : "");
+	return do_update_pr(mapname, "setprkey", str);
 }
Signature · 6 lines
2.17.2

--
dm-devel mailing list
[email protected]
https://listman.redhat.com/mailman/listinfo/dm-devel
Reviewed-by on v1 2/2
#2 lore raw

Re: [dm-devel] [PATCH 2/2] libmpathpersist: fix command keyword ordering

Quoted 11 lines · Benjamin Marzinski

When libmpathpersist was communicating with multipathd, it wasn't using
the correct keyword order for the commands, as specified in the CLI commands reference. Since commit f812466f, multipathd requires commands
to be ordered correctly. Fix the ordering.

Fixes f812466f ("multipathd: more robust command parsing")
Cc lixiaokeng <[email protected]>
Quoted 54 lines

---
 libmpathpersist/mpath_updatepr.c | 23 ++++++++++++-----------

 1 file changed, 12 insertions(+), 11 deletions(-)
libmpathpersist/mpath_updatepr.c +0 -0
index 5824c169..4529a82b 100644
@@ -18,7 +18,7 @@
  #include "mpathpr.h"


libmpathpersist/mpath_updatepr.c +1 -1
-static int do_update_pr(char *alias, char *arg)
+static int do_update_pr(char *alias, char *cmd, char *key)

 {
        int fd;
        char str[256];

libmpathpersist/mpath_updatepr.c +0 -0
@@ -31,7 +31,10 @@ static int do_update_pr(char *alias, char *arg)
                 return -1;
         }


-       snprintf(str,sizeof(str),"map %s %s", alias, arg); +       if (key)
+               snprintf(str,sizeof(str),"%s map %s key %s", cmd, alias, key);
+       else
+               snprintf(str,sizeof(str),"%s map %s", cmd, alias); condlog (2, "%s: pr message=%s", alias, str); if (send_packet(fd, str) != 0) {
                condlog(2, "%s: message=%s send error=%d", alias, str, errno);

libmpathpersist/mpath_updatepr.c +0 -0
@@ -56,18 +59,16 @@ static int do_update_pr(char *alias, char *arg)
  }


 int update_prflag(char *mapname, int set) {
-       return do_update_pr(mapname, (set)? "setprstatus" :
"unsetprstatus");
+       return do_update_pr(mapname, (set)? "setprstatus" :
"unsetprstatus",
+                           NULL);
 }

 int update_prkey_flags(char *mapname, uint64_t prkey, uint8_t sa_flags) {
        char str[256];
-       char *flagstr = "";

-       if (sa_flags & MPATH_F_APTPL_MASK)
-               flagstr = ":aptpl";
-       if (prkey)
-               sprintf(str, "setprkey key %" PRIx64 "%s", prkey, flagstr);
-       else

Quoted 8 lines

-               sprintf(str, "unsetprkey");
-       return do_update_pr(mapname, str);
+       if (!prkey)
+               return do_update_pr(mapname, "unsetprkey", NULL); +       sprintf(str, "%" PRIx64 "%s", prkey,
+               (sa_flags & MPATH_F_APTPL_MASK) ? ":aptpl" : ""); +       return do_update_pr(mapname, "setprkey", str); }

--
dm-devel mailing list
[email protected]
https://listman.redhat.com/mailman/listinfo/dm-devel

Reviewed-by on v1 1/2
#3 lore raw
In reply to Benjamin Marzinski
Quoted 12 lines · Benjamin Marzinski

The multpath-tools do not support the io-affinity path selector.  We always add a repeat count as the path argument. The io-affinity selector
doesn't take one. Instead it takes a bitmap of CPUs that a path can run
on. This isn't something that lends itself to the kind of auto-assembling that multipathd does. But even if we did want to try to
support this path-selector, until we do, we shouldn't be listing it in
the multipath.conf documentation.

Quoted 1 line

--
dm-devel mailing list
[email protected]
https://listman.redhat.com/mailman/listinfo/dm-devel

#4 lore raw
In reply to Benjamin Marzinski
Quoted 19 lines · bmarzins at redhat.com (Benjamin Marzinski)

The multpath-tools do not support the io-affinity path selector. We always add a repeat count as the path argument. The io-affinity selector doesn't take one. Instead it takes a bitmap of CPUs that a path can run on. This isn't something that lends itself to the kind of auto-assembling that multipathd does. But even if we did want to try to support this path-selector, until we do, we shouldn't be listing it in the multipath.conf documentation.

Signed-off-by Benjamin Marzinski <bmarzins at redhat.com>
---
  multipath/multipath.conf.5 | 4 ----
  1 file changed, 4 deletions(-)
multipath/multipath.conf.5 +0 -0
index 1fea9d5a..3a45ac89 100644

multipath/multipath.conf.5 of outstanding I/O to the path and its relative throughput. · line 210

Quoted 10 lines
multipath/multipath.conf.5 +0 -4
@@ -205,10 +205,6 @@ of outstanding I/O to the path and its relative throughput.
  estimation of future service time based on the history of previous I/O submitted
  to each path.
  .TP
-.I "io-affinity 0"
-(Since 5.11 kernel) Choose the path for the next bunch of I/O based on a CPU to
-path mapping the user passes in and what CPU we are executing on.
-.TP
  The default is: \fBservice-time 0\fR
  .RE
  .

I think the main, and only?, consumer for this path selector is Exadata.

--
dm-devel mailing list
[email protected]
https://listman.redhat.com/mailman/listinfo/dm-devel

#5 lore raw
Quoted 25 lines · Xose Vazquez Perez
Quoted 23 lines · bmarzins at redhat.com (Benjamin Marzinski)

The multpath-tools do not support the io-affinity path selector.  We always add a repeat count as the path argument. The io-affinity selector doesn't take one. Instead it takes a bitmap of CPUs that a path can run on. This isn't something that lends itself to the kind of auto-assembling that multipathd does. But even if we did want to try to support this path-selector, until we do, we shouldn't be listing it in the multipath.conf documentation.

Signed-off-by Benjamin Marzinski <bmarzins at redhat.com>

---
  multipath/multipath.conf.5 | 4 ----

  1 file changed, 4 deletions(-)
multipath/multipath.conf.5 +0 -0
index 1fea9d5a..3a45ac89 100644
@@ -205,10 +205,6 @@ of outstanding I/O to the path and its relative throughput.

  estimation of future service time based on the history of previous I/O submitted to each path.
  .TP
-.I "io-affinity 0"

Quoted 8 lines
Quoted 6 lines
multipath/multipath.conf.5 +0 -3
-(Since 5.11 kernel) Choose the path for the next bunch of I/O based on a CPU to
-path mapping the user passes in and what CPU we are executing on.
-.TP

  The default is: \fBservice-time 0\fR
  .RE
  .

I think the main, and only?, consumer for this path selector is Exadata.

Not exadata but another user in oracle.

The patch is ok with me. We've been setting it up manually. When I get time to do multipathd support I'll update the docs as well.

--
dm-devel mailing list
[email protected]
https://listman.redhat.com/mailman/listinfo/dm-devel