OCFS Configurable timeouts - Revision 2
cover OCFS Configurable timeouts - Revision 2
The following is an update to the 3rd patch in the configurable timeouts series.
Changes to the original patch include:
- setting and checking of o2hb_heartbeat_timeout_ms in handshakes
- better logic in o2net_advance_rx() removing the need for dummy fields in o2net_msg struct
- tracks how many peers are connected
- prevents o2net_idle_timeout_ms and o2net_keepalive_delay_ms from being modified when we have connected peers
The rationale for including these changes in the final patch is that patch 2 creates the capability and patch 3 ensures it it used sanely across all nodes.
Still to come is a patch which makes use of the to_o2nm_cluster_from_node() function as previously suggested by Mark.
Andrew
--
PATCH 1/1 ocfs2-timeout-protocol.patch
From: Andrew Beekhof <[email protected]>
Subject: [patch 1/1] OCFS2 Configurable timeouts - Protocol changes
Modify the OCFS2 handshake to ensure essential timeouts are configured
identically on all nodes.
Included is the field for userspace-heartbeat timeout to avoid the need for
further protocol changes.
The addition of two dummy fields is a temporary measure to
satisfy the logic in o2net_check_handshake() and will be rectified in a future version of this patch
--- fs/ocfs2/cluster/nodemanager.c | 19 ++++++- fs/ocfs2/cluster/tcp.c | 102 ++++++++++++++++++++++++++++++++++------ fs/ocfs2/cluster/tcp.h | 1 fs/ocfs2/cluster/tcp_internal.h | 6 +- 4 files changed, 111 insertions(+), 17 deletions(-)
ocfs2/cluster/nodemanager.c renamed from ocfs2/cluster/nodemanager.c.orig +17 -2
@@ -574,7 +574,14 @@ static ssize_t o2nm_cluster_attr_idle_ti ret =o2nm_cluster_attr_write (page, count, &val);if (ret >0 ) { -if (val <= cluster->cl_keepalive_delay_ms) { +if (cluster->cl_idle_timeout_ms != val + &&o2net_num_connected_peers ()) { +mlog (ML_NOTICE ,"o2net: cannot change idle timeout after " +"the first peer has agreed to it. %d connected peers\n" , +o2net_num_connected_peers ()); +return -EINVAL ; + + }else if (val <= cluster->cl_keepalive_delay_ms) {mlog (ML_NOTICE ,"o2net: idle timeout must be larger " "than keepalive delay\n" );return -EINVAL ;
@@ -601,7 +608,15 @@ static ssize_t o2nm_cluster_attr_keepali ret =o2nm_cluster_attr_write (page, count, &val);if (ret >0 ) { -if (val >= cluster->cl_idle_timeout_ms) { + +if (cluster->cl_keepalive_delay_ms != val + &&o2net_num_connected_peers ()) { +mlog (ML_NOTICE ,"o2net: cannot change keepalive delay after " +"the first peer has agreed to it. %d connected peers\n" , +o2net_num_connected_peers ()); +return -EINVAL ; + + }else if (val >= cluster->cl_idle_timeout_ms) {mlog (ML_NOTICE ,"o2net: keepalive delay must be " "smaller than idle timeout\n" );return -EINVAL ;
ocfs2/cluster/tcp.c renamed from ocfs2/cluster/tcp.c.orig +88 -14
170 lines not on this page · raw message
ocfs2/cluster/tcp.h renamed from ocfs2/cluster/tcp.h.orig +1 -0
@@ -108,6 +108,7 @@ void o2net_unregister_hb_callbacks(void)int o2net_start_listening (struct o2nm_node *node);void o2net_stop_listening (struct o2nm_node *node);void o2net_disconnect_node (struct o2nm_node *node); +int o2net_num_connected_peers (void );int o2net_init (void );void o2net_exit (void );
ocfs2/cluster/tcp_internal.h renamed from ocfs2/cluster/tcp_internal.h.orig +5 -1
@@ -48,10 +48,14 @@* - full 64 bit i_size in the metadata lock lvbs * - introduction of "rw" lock and pushing meta/data locking down */ -#define O2NET_PROTOCOL_VERSION 4ULL +#define O2NET_PROTOCOL_VERSION 5ULLstruct o2net_handshake {__be64 protocol_version;__be64 connector_id; +__be32 o2hb_heartbeat_timeout_ms; +__be32 o2net_idle_timeout_ms; +__be32 o2net_keepalive_delay_ms; +__be32 o2net_reconnect_delay_ms; };struct o2net_node {
--