cman: Fix udpb handling in cman-preconfig
PATCH 1/1 cman: Fix udpb handling in cman-preconfig
--- cman/daemon/cman-preconfig.c | 2 +- 1 files changed, 1 insertions(+), 1 deletions(-)
cman/daemon/cman-preconfig.c +1 -1
@@ -286,7 +286,7 @@ static int add_ifaddr(struct objdb_iface_ver0 *objdb, char *mcast, char *ifaddr,int ret =0 ;const char *tx_mech_to_str[] = { [TX_MECH_UDP ] ="udp" , - [TX_MECH_UDPB ] ="udp" , + [TX_MECH_UDPB ] ="udpb" , [TX_MECH_UDPU ] ="udpu" , [TX_MECH_RDMA ] ="iba" , };
Signature · 1 line
1.7.2.3
PATCH 1/1 cman: Fix udpb handling in cman-preconfig
Lon,
even it may "logically" seemed that original code is incorrect and your patch make sense, it's not the case.
Actually, corosync has 3 transports:
udp, udpu and iba.
Broadcast is same transport as udp but only has broadcast flag set on.
In other words, this patch is not correct.
Regards,
Honza
Quoted 18 lines · Lon Hohberger
Signed-off-by Lon Hohberger <[email protected]>--- cman/daemon/cman-preconfig.c | 2 +- 1 files changed, 1 insertions(+), 1 deletions(-)cman/daemon/cman-preconfig.c +1
-1@@ -286,7 +286,7 @@ static int add_ifaddr(struct objdb_iface_ver0 *objdb, char *mcast, char *ifaddr,int ret =0 ;const char *tx_mech_to_str[] = { [TX_MECH_UDP ] ="udp" , - [TX_MECH_UDPB ] ="udp" , + [TX_MECH_UDPB ] ="udpb" , [TX_MECH_UDPU ] ="udpu" , [TX_MECH_RDMA ] ="iba" , };
PATCH 1/1 cman: Fix udpb handling in cman-preconfig
Quoted 11 lines · Jan Friesse
Lon,
even it may "logically" seemed that original code is incorrect and your patch make sense, it's not the case.Actually, corosync has 3 transports:
udp, udpu and iba.Broadcast is same transport as udp but only has broadcast flag set on.
In other words, this patch is not correct.
So in order to configure broadcast we still need to use:
<cman transport="udp" broadcast="yes"/> ?
I find that redundant tho...
Cheers
Fabio
Quoted 23 lines
Regards,
HonzaQuoted 18 lines · Lon Hohberger
Signed-off-by Lon Hohberger <[email protected]>--- cman/daemon/cman-preconfig.c | 2 +- 1 files changed, 1 insertions(+), 1 deletions(-)cman/daemon/cman-preconfig.c +1
-1@@ -286,7 +286,7 @@ static int add_ifaddr(struct objdb_iface_ver0 *objdb, char *mcast, char *ifaddr,int ret =0 ;const char *tx_mech_to_str[] = { [TX_MECH_UDP ] ="udp" , - [TX_MECH_UDPB ] ="udp" , + [TX_MECH_UDPB ] ="udpb" , [TX_MECH_UDPU ] ="udpu" , [TX_MECH_RDMA ] ="iba" , };
PATCH 1/1 cman: Fix udpb handling in cman-preconfig
Quoted 13 lines · Fabio M. Di Nitto
Quoted 11 lines · Jan Friesse
Lon,
even it may "logically" seemed that original code is incorrect and your patch make sense, it's not the case.Actually, corosync has 3 transports:
udp, udpu and iba.Broadcast is same transport as udp but only has broadcast flag set on.
In other words, this patch is not correct.
Quoted 9 lines
So in order to configure broadcast we still need to use:
<cman transport="udp" broadcast="yes"/> ?
I find that redundant tho...
Cheers
Fabio
No.
Actually. Original idea was to use transport keyword and give user ability to use udp (for multicast), udpu (for unicast), udpb (for broadcast) and rdma. But also we must keep compatibility so thats why broadcast keyword still exists. So actually, transport="udpb" is same as transport="udp" broadcast="yes".
Regards,
Honza
Quoted 23 lines
Quoted 22 lines
Regards,
HonzaQuoted 18 lines · Lon Hohberger
Signed-off-by Lon Hohberger <[email protected]>--- cman/daemon/cman-preconfig.c | 2 +- 1 files changed, 1 insertions(+), 1 deletions(-)cman/daemon/cman-preconfig.c +1
-1@@ -286,7 +286,7 @@ static int add_ifaddr(struct objdb_iface_ver0 *objdb, char *mcast, char *ifaddr,int ret =0 ;const char *tx_mech_to_str[] = { [TX_MECH_UDP ] ="udp" , - [TX_MECH_UDPB ] ="udp" , + [TX_MECH_UDPB ] ="udpb" , [TX_MECH_UDPU ] ="udpu" , [TX_MECH_RDMA ] ="iba" , };
PATCH 1/1 cman: Fix udpb handling in cman-preconfig
In reply to Jan FriesseQuoted 10 lines · Jan Friesse
Lon,
even it may "logically" seemed that original code is incorrect and your patch make sense, it's not the case.Actually, corosync has 3 transports:
udp, udpu and iba.Broadcast is same transport as udp but only has broadcast flag set on.
In other words, this patch is not correct.
Ok, reverted. Sorry for clouding the issue,
-- Lon