sin/samba-mirror
1
0
Fork 0
mirror of https://github.com/samba-team/samba.git synced 2025-01-03 01:18:10 +03:00
Code

ctdb-tool: Add UNKNOWN pseudo state

Browse Source 
When a node is starting, CTDB reports remote nodes as unhealthy by
default.  This can be misleading.

To hide this, report an "UNKNOWN" pseudo state when a remote node is
not disconnected and the runstate is less than or equal to
"FIRST_RECOVERY".

Signed-off-by: Vinit Agnihotri <vagnihotri@ddn.com>
Signed-off-by: Martin Schwenke <martin@meltin.net>
Reviewed-by: Amitay Isaacs <amitay@gmail.com>
This commit is contained in:
Vinit Agnihotri 2022-04-26 17:20:21 +10:00 committed by Amitay Isaacs
parent 428bc71f98
commit 794f125802
12 changed files with 122 additions and 47 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

8
ctdb/tests/UNIT/tool/ctdb.nodestatus.001.sh
View File

@ -25,9 +25,9 @@ EOF
simple_test all
required_result 0 <<EOF
|Node|IP|Disconnected|Banned|Disabled|Unhealthy|Stopped|Inactive|PartiallyOnline|ThisNode|
|0|192.168.20.41|0|0|0|0|0|0|0|N|
|1|192.168.20.42|0|0|0|0|0|0|0|N|
|2|192.168.20.43|0|0|0|0|0|0|0|Y|
|Node|IP|Disconnected|Unknown|Banned|Disabled|Unhealthy|Stopped|Inactive|PartiallyOnline|ThisNode|
|0|192.168.20.41|0|0|0|0|0|0|0|0|N|
|1|192.168.20.42|0|0|0|0|0|0|0|0|N|
|2|192.168.20.43|0|0|0|0|0|0|0|0|Y|
EOF
simple_test -X all

8
ctdb/tests/UNIT/tool/ctdb.nodestatus.002.sh
View File

@ -25,9 +25,9 @@ EOF
simple_test all
required_result 1 <<EOF
|Node|IP|Disconnected|Banned|Disabled|Unhealthy|Stopped|Inactive|PartiallyOnline|ThisNode|
|0|192.168.20.41|0|0|0|0|0|0|0|N|
|1|192.168.20.42|1|0|0|0|0|1|0|N|
|2|192.168.20.43|0|0|0|0|0|0|0|Y|
|Node|IP|Disconnected|Unknown|Banned|Disabled|Unhealthy|Stopped|Inactive|PartiallyOnline|ThisNode|
|0|192.168.20.41|0|0|0|0|0|0|0|0|N|
|1|192.168.20.42|1|0|0|0|0|0|1|0|N|
|2|192.168.20.43|0|0|0|0|0|0|0|0|Y|
EOF
simple_test -X all

8
ctdb/tests/UNIT/tool/ctdb.nodestatus.003.sh
View File

@ -25,9 +25,9 @@ EOF
simple_test all
required_result 2 <<EOF
|Node|IP|Disconnected|Banned|Disabled|Unhealthy|Stopped|Inactive|PartiallyOnline|ThisNode|
|0|192.168.20.41|0|0|0|1|0|0|0|N|
|1|192.168.20.42|0|0|0|0|0|0|0|N|
|2|192.168.20.43|0|0|0|0|0|0|0|Y|
|Node|IP|Disconnected|Unknown|Banned|Disabled|Unhealthy|Stopped|Inactive|PartiallyOnline|ThisNode|
|0|192.168.20.41|0|0|0|0|1|0|0|0|N|
|1|192.168.20.42|0|0|0|0|0|0|0|0|N|
|2|192.168.20.43|0|0|0|0|0|0|0|0|Y|
EOF
simple_test -X all

4
ctdb/tests/UNIT/tool/ctdb.nodestatus.004.sh
View File

@ -22,7 +22,7 @@ EOF
simple_test
required_result 0 <<EOF
|Node|IP|Disconnected|Banned|Disabled|Unhealthy|Stopped|Inactive|PartiallyOnline|ThisNode|
|2|192.168.20.43|0|0|0|0|0|0|0|Y|
|Node|IP|Disconnected|Unknown|Banned|Disabled|Unhealthy|Stopped|Inactive|PartiallyOnline|ThisNode|
|2|192.168.20.43|0|0|0|0|0|0|0|0|Y|
EOF
simple_test -X

4
ctdb/tests/UNIT/tool/ctdb.nodestatus.005.sh
View File

@ -22,7 +22,7 @@ EOF
simple_test 0
required_result 2 <<EOF
|Node|IP|Disconnected|Banned|Disabled|Unhealthy|Stopped|Inactive|PartiallyOnline|ThisNode|
|0|192.168.20.41|0|0|0|1|0|0|0|N|
|Node|IP|Disconnected|Unknown|Banned|Disabled|Unhealthy|Stopped|Inactive|PartiallyOnline|ThisNode|
|0|192.168.20.41|0|0|0|0|1|0|0|0|N|
EOF
simple_test -X 0

4
ctdb/tests/UNIT/tool/ctdb.nodestatus.006.sh
View File

@ -22,8 +22,8 @@ EOF
simple_test 0
required_result 36 <<EOF
|Node|IP|Disconnected|Banned|Disabled|Unhealthy|Stopped|Inactive|PartiallyOnline|ThisNode|
|0|192.168.20.41|0|0|1|0|1|1|0|N|
|Node|IP|Disconnected|Unknown|Banned|Disabled|Unhealthy|Stopped|Inactive|PartiallyOnline|ThisNode|
|0|192.168.20.41|0|0|0|1|0|1|1|0|N|
EOF
simple_test -X 0

8
ctdb/tests/UNIT/tool/ctdb.status.001.sh
View File

@ -38,9 +38,9 @@ EOF
simple_test
required_result 0 <<EOF
|Node|IP|Disconnected|Banned|Disabled|Unhealthy|Stopped|Inactive|PartiallyOnline|ThisNode|
|0|192.168.20.41|0|0|0|0|0|0|0|Y|
|1|192.168.20.42|0|0|0|0|0|0|0|N|
|2|192.168.20.43|0|0|0|0|0|0|0|N|
|Node|IP|Disconnected|Unknown|Banned|Disabled|Unhealthy|Stopped|Inactive|PartiallyOnline|ThisNode|
|0|192.168.20.41|0|0|0|0|0|0|0|0|Y|
|1|192.168.20.42|0|0|0|0|0|0|0|0|N|
|2|192.168.20.43|0|0|0|0|0|0|0|0|N|
EOF
simple_test -X

8
ctdb/tests/UNIT/tool/ctdb.status.002.sh
View File

@ -38,9 +38,9 @@ EOF
simple_test
required_result 0 <<EOF
|Node|IP|Disconnected|Banned|Disabled|Unhealthy|Stopped|Inactive|PartiallyOnline|ThisNode|
|0|192.168.20.41|0|0|0|1|0|0|0|N|
|1|192.168.20.42|0|0|0|0|0|0|0|Y|
|2|192.168.20.43|0|0|0|0|0|0|0|N|
|Node|IP|Disconnected|Unknown|Banned|Disabled|Unhealthy|Stopped|Inactive|PartiallyOnline|ThisNode|
|0|192.168.20.41|0|0|0|0|1|0|0|0|N|
|1|192.168.20.42|0|0|0|0|0|0|0|0|Y|
|2|192.168.20.43|0|0|0|0|0|0|0|0|N|
EOF
simple_test -X

16
ctdb/tests/scripts/integration.bash
View File

@ -446,16 +446,16 @@ node_has_status ()
local bits
case "$status" in
unhealthy) bits="?|?|?|1|*" ;;
healthy) bits="?|?|?|0|*" ;;
unhealthy) bits="?|?|?|?|1|*" ;;
healthy) bits="?|?|?|?|0|*" ;;
disconnected) bits="1|*" ;;
connected) bits="0|*" ;;
banned) bits="?|1|*" ;;
unbanned) bits="?|0|*" ;;
disabled) bits="?|?|1|*" ;;
enabled) bits="?|?|0|*" ;;
stopped) bits="?|?|?|?|1|*" ;;
notstopped) bits="?|?|?|?|0|*" ;;
banned) bits="?|?|1|*" ;;
unbanned) bits="?|?|0|*" ;;
disabled) bits="?|?|?|1|*" ;;
enabled) bits="?|?|?|0|*" ;;
stopped) bits="?|?|?|?|?|1|*" ;;
notstopped) bits="?|?|?|?|?|0|*" ;;
*)
echo "node_has_status: unknown status \"$status\""
return 1

83
ctdb/tools/ctdb.c
View File

@ -52,6 +52,8 @@
#define SRVID_CTDB_TOOL (CTDB_SRVID_TOOL_RANGE | 0x0001000000000000LL)
#define SRVID_CTDB_PUSHDB (CTDB_SRVID_TOOL_RANGE | 0x0002000000000000LL)
#define NODE_FLAGS_UNKNOWN 0x00000040
static struct {
const char *debuglevelstr;
int timelimit;
@ -111,6 +113,7 @@ static const char *pretty_print_flags(TALLOC_CTX *mem_ctx, uint32_t flags)
const char *name;
} flag_names[] = {
{ NODE_FLAGS_DISCONNECTED, "DISCONNECTED" },
{ NODE_FLAGS_UNKNOWN, "UNKNOWN" },
{ NODE_FLAGS_PERMANENTLY_DISABLED, "DISABLED" },
{ NODE_FLAGS_BANNED, "BANNED" },
{ NODE_FLAGS_UNHEALTHY, "UNHEALTHY" },
@ -367,6 +370,64 @@ done:
return true;
}
/*
* Remote nodes are initialised as UNHEALTHY in the daemon and their
* true status is udpated after they are connected. However, there
* is a small window when a healthy node may be shown as unhealthy
* between connecting and the status update. Hide this for nodes
* that are not DISCONNECTED nodes by reporting them as UNKNOWN until
* the runstate passes FIRST_RECOVERY. Code paths where this is used
* do not make any control decisions depending upon unknown/unhealthy
* state.
*/
static struct ctdb_node_map *get_nodemap_unknown(
TALLOC_CTX *mem_ctx,
struct ctdb_context *ctdb,
struct ctdb_node_map *nodemap_in)
{
unsigned int i;
int ret;
enum ctdb_runstate runstate;
struct ctdb_node_map *nodemap;
ret = ctdb_ctrl_get_runstate(mem_ctx,
ctdb->ev,
ctdb->client,
ctdb->cmd_pnn,
TIMEOUT(),
&runstate);
if (ret != 0 ) {
printf("Unable to get runstate");
return NULL;
}
nodemap = talloc_nodemap(mem_ctx, nodemap_in);
if (nodemap == NULL) {
printf("Unable to get nodemap");
return NULL;
}
nodemap->num = nodemap_in->num;
for (i=0; i<nodemap->num; i++) {
struct ctdb_node_and_flags *node_in = &nodemap_in->node[i];
struct ctdb_node_and_flags *node = &nodemap->node[i];
*node = *node_in;
if (node->flags & NODE_FLAGS_DELETED) {
continue;
}
if ((runstate <= CTDB_RUNSTATE_FIRST_RECOVERY) &&
!(node->flags & NODE_FLAGS_DISCONNECTED) &&
(node->pnn != ctdb->cmd_pnn)) {
node->flags = NODE_FLAGS_UNKNOWN;
}
}
return nodemap;
}
/* Compare IP address */
static bool ctdb_same_ip(ctdb_sock_addr *ip1, ctdb_sock_addr *ip2)
{
@ -826,11 +887,12 @@ static void print_nodemap_machine(TALLOC_CTX *mem_ctx,
struct ctdb_node_and_flags *node;
unsigned int i;
printf("%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s\n",
printf("%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s\n",
options.sep,
"Node", options.sep,
"IP", options.sep,
"Disconnected", options.sep,
"Unknown", options.sep,
"Banned", options.sep,
"Disabled", options.sep,
"Unhealthy", options.sep,
@ -845,12 +907,13 @@ static void print_nodemap_machine(TALLOC_CTX *mem_ctx,
continue;
}
printf("%s%u%s%s%s%d%s%d%s%d%s%d%s%d%s%d%s%d%s%c%s\n",
printf("%s%u%s%s%s%d%s%d%s%d%s%d%s%d%s%d%s%d%s%d%s%c%s\n",
options.sep,
node->pnn, options.sep,
ctdb_sock_addr_to_string(mem_ctx, &node->addr, false),
options.sep,
!! (node->flags & NODE_FLAGS_DISCONNECTED), options.sep,
!! (node->flags & NODE_FLAGS_UNKNOWN), options.sep,
!! (node->flags & NODE_FLAGS_BANNED), options.sep,
!! (node->flags & NODE_FLAGS_PERMANENTLY_DISABLED),
options.sep,
@ -935,6 +998,7 @@ static void print_status(TALLOC_CTX *mem_ctx,
static int control_status(TALLOC_CTX *mem_ctx, struct ctdb_context *ctdb,
int argc, const char **argv)
{
struct ctdb_node_map *nodemap_in;
struct ctdb_node_map *nodemap;
struct ctdb_vnn_map *vnnmap;
int recmode;
@ -945,7 +1009,12 @@ static int control_status(TALLOC_CTX *mem_ctx, struct ctdb_context *ctdb,
usage("status");
}
nodemap = get_nodemap(ctdb, false);
nodemap_in = get_nodemap(ctdb, false);
if (nodemap_in == NULL) {
return 1;
}
nodemap = get_nodemap_unknown(mem_ctx, ctdb, nodemap_in);
if (nodemap == NULL) {
return 1;
}
@ -5603,6 +5672,7 @@ static int control_nodestatus(TALLOC_CTX *mem_ctx, struct ctdb_context *ctdb,
int argc, const char **argv)
{
const char *nodestring = NULL;
struct ctdb_node_map *nodemap_in;
struct ctdb_node_map *nodemap;
unsigned int i;
int ret;
@ -5619,7 +5689,12 @@ static int control_nodestatus(TALLOC_CTX *mem_ctx, struct ctdb_context *ctdb,
}
}
if (! parse_nodestring(mem_ctx, ctdb, nodestring, &nodemap)) {
if (! parse_nodestring(mem_ctx, ctdb, nodestring, &nodemap_in)) {
return 1;
}
nodemap = get_nodemap_unknown(mem_ctx, ctdb, nodemap_in);
if (nodemap == NULL) {
return 1;
}

8
ctdb/tools/ctdb_lvs
View File

@ -32,7 +32,7 @@ EOF
nodestatus_X=""
# Fields are:
# Node|IP|Disconnected|Banned|Disabled|Unhealthy|Stopped|Inactive|PartiallyOnline|ThisNode
# Node|IP|Disconnected|Unknown|Banned|Disabled|Unhealthy|Stopped|Inactive|PartiallyOnline|ThisNode
get_nodestatus_X ()
{
# Result is cached in global variable nodestatus_X
@ -100,11 +100,11 @@ filter_nodes ()
# them, so the first to succeed will print the nodes.
# First try for a fully active and healthy node, so must not
# be DISABLED, UNHEALTHY or INACTIVE (last covers
# be UNKNOWN, DISABLED, UNHEALTHY or INACTIVE (last covers
# DISCONNECTED, BANNED or STOPPED)
awk -F '|' -v ns="$_ns" '
BEGIN { ret = 255 }
ns ~ "@" $2 "@" && $5 == 0 && $6 == 0 && $8 == 0 {
ns ~ "@" $2 "@" && $4 == 0 && $6 == 0 && $7 == 0 && $9 == 0 {
print $1, $2 ; ret=0
}
END { exit ret }
@ -115,7 +115,7 @@ EOF
# DISABLED
awk -F '|' -v ns="$_ns" '
BEGIN { ret = 255 }
ns ~ "@" $2 "@" && $5 == 0 && $8 == 0 {
ns ~ "@" $2 "@" && $6 == 0 && $9 == 0 {
print $1, $2 ; ret=0
}
END { exit ret }

10
ctdb/tools/ctdb_natgw
View File

@ -32,7 +32,7 @@ EOF
nodestatus_X=""
# Fields are:
# Node|IP|Disconnected|Banned|Disabled|Unhealthy|Stopped|Inactive|PartiallyOnline|ThisNode
# Node|IP|Disconnected|Unknown|Banned|Disabled|Unhealthy|Stopped|Inactive|PartiallyOnline|ThisNode
get_nodestatus_X ()
{
# Result is cached in global variable nodestatus_X
@ -102,12 +102,12 @@ EOF
# the first to succeed will select the leader node.
# First try for a fully active and healthy node, so must not be
# DISABLED, UNHEALTHY or INACTIVE (last covers DISCONNECTED,
# UNKNOWN, DISABLED, UNHEALTHY or INACTIVE (last covers DISCONNECTED,
# BANNED or STOPPED)
awk -F '|' -v ms="$_ms" \
'BEGIN { ret = 2 }
ms ~ "@" $2 "@" &&
$5 == 0 && $6 == 0 && $8 == 0 { print $1, $2 ; ret=0 ; exit }
$4 == 0 && $6 == 0 && $7 == 0 && $9 == 0 { print $1, $2 ; ret=0 ; exit }
END { exit ret }' <<EOF ||
$nodestatus_X
EOF
@ -116,7 +116,7 @@ EOF
awk -F '|' -v ms="$_ms" \
'BEGIN { ret = 2 }
ms ~ "@" $2 "@" &&
$3 == 0 && $5 == 0 && $7 == 0 { print $1, $2 ; ret=0 ; exit }
$3 == 0 && $6 == 0 && $8 == 0 { print $1, $2 ; ret=0 ; exit }
END { exit ret }' <<EOF ||
$nodestatus_X
EOF
@ -125,7 +125,7 @@ EOF
awk -F '|' -v ms="$_ms" \
'BEGIN { ret = 2 }
ms ~ "@" $2 "@" &&
$3 == 0 && $5 == 0 { print $1, $2 ; ret=0 ; exit }
$3 == 0 && $6 == 0 { print $1, $2 ; ret=0 ; exit }
END { exit ret }' <<EOF
$nodestatus_X
EOF