improve activation monitoring option processing
PATCH v2 1/1 improve activation monitoring option processing
Add "monitoring" option to "activation" section of lvm.conf
Have clvmd consult the lvm.conf "activation/monitoring" too.
Introduce toollib.c:get_activation_monitoring_mode().
Error out when both --monitor and --ignoremonitoring are provided.
Add --monitor and --ignoremonitoring support to lvcreate. Update lvcreate man page accordingly.
Clarify that '--monitor' controls the start and stop of monitoring in the {vg,lv}change man pages.
--- daemons/clvmd/lvm-functions.c | 8 ++++++-- doc/example.conf | 4 ++++ lib/metadata/lv_manip.c | 2 ++ lib/metadata/metadata-exported.h | 1 + liblvm/lvm_lv.c | 1 + man/lvchange.8.in | 3 +-- man/lvcreate.8.in | 14 ++++++++++++++ man/vgchange.8.in | 3 +-- tools/commands.h | 14 +++++++++----- tools/lvchange.c | 9 +++++---- tools/lvcreate.c | 3 +++ tools/toollib.c | 21 +++++++++++++++++++++ tools/toollib.h | 3 +++ tools/vgchange.c | 9 +++++---- 14 files changed, 76 insertions(+), 19 deletions(-)
daemons/clvmd/lvm-functions.c +6 -2
@@ -542,8 +542,12 @@ int do_lock_lv(unsigned char command, unsigned char lock_flags, char *resource)if (lock_flags &LCK_MIRROR_NOSYNC_MODE )init_mirror_in_sync (0 ); -if (!(lock_flags &LCK_DMEVENTD_MONITOR_MODE )) -init_dmeventd_monitor (DEFAULT_DMEVENTD_MONITOR ); +if (!(lock_flags &LCK_DMEVENTD_MONITOR_MODE )) { +int dmeventd_mode = +find_config_tree_bool (cmd,"activation/monitoring" , +DEFAULT_DMEVENTD_MONITOR ); +init_dmeventd_monitor (dmeventd_mode); + } cmd->partial_activation =0 ;
doc/example.conf +4 -0
@@ -428,6 +428,10 @@ activation { # which used mlockall() to pin the whole process's memory while activating # devices. # use_mlockall = 0 + + # Monitoring is enabled by default when activating logical volumes. + # Set to 0 to disable monitoring or use the --ignoremonitoring option. + # monitoring = 1 }
lib/metadata/lv_manip.c +2 -0
@@ -3056,6 +3056,8 @@ int lv_create_single(struct volume_group *vg,backup (vg); +init_dmeventd_monitor (lp->activation_monitoring); +if (lp->snapshot) {if (!activate_lv_excl (cmd, lv)) { log_error("Aborting. Failed to activate snapshot "
lib/metadata/metadata-exported.h +1 -0
@@ -535,6 +535,7 @@ struct lvcreate_params {int minor;/* all */ int log_count;/* mirror */ int nosync;/* mirror */ +int activation_monitoring;/* all */ char *origin;/* snap */ const char *vg_name;/* all */
liblvm/lvm_lv.c +1 -0
@@ -111,6 +111,7 @@ static void _lv_set_default_params(struct lvcreate_params *lp, lp->zero =1 ; lp->major =-1 ; lp->minor =-1 ; + lp->activation_monitoring =DEFAULT_DMEVENTD_MONITOR ; lp->vg_name = vg->name; lp->lv_name = lvname;/* FIXME: check this for safety */ lp->pvh = &vg->pvs;
man/lvchange.8.in +1 -2
@@ -56,8 +56,7 @@ of your data. Set the minor number. .TP .I \-\-monitor y|n -Controls whether or not a mirrored logical volume is monitored by -dmeventd, if it is installed. +Start or stop monitoring a logical volume, if dmeventd is installed. If a device used by a monitored mirror reports an I/O error, the failure is handled according to \fBmirror_image_fault_policy\fP and \fBmirror_log_fault_policy\fP
man/lvcreate.8.in +14 -0
@@ -7,6 +7,8 @@ lvcreate \- create a logical volume in an existing volume group [\-\-alloc AllocationPolicy] [\-A|\-\-autobackup y|n] [\-C|\-\-contiguous y|n] [\-d|\-\-debug] [\-h|\-?|\-\-help] [\-\-noudevsync] +[\-\-ignoremonitoring] +[\-\-monitor {y|n}] [\-i|\-\-stripes Stripes [\-I|\-\-stripesize StripeSize]] {\-l|\-\-extents LogicalExtentsNumber[%{VG|PVS|FREE}] | \-L|\-\-size LogicalVolumeSize[bBsSkKmMgGtTpPeE]}
@@ -26,6 +28,8 @@ VolumeGroupName [PhysicalVolumePath[:PE[-PE]]...] \-L|\-\-size LogicalVolumeSize[bBsSkKmMgGtTpPeE]} [\-c|\-\-chunksize ChunkSize] [\-\-noudevsync] +[\-\-ignoremonitoring] +[\-\-monitor {y|n}] \-n|\-\-name SnapshotLogicalVolumeName {{\-s|\-\-snapshot} OriginalLogicalVolumePath |
@@ -128,6 +132,16 @@ process will not wait for notification from udev. It will continue irrespective of any possible udev processing in the background. You should only use this if udev is not running or has rules that ignore the devices LVM2 creates. +.I \-\-monitor y|n +Start or avoid monitoring a logical volume, if dmeventd is installed. +If a device used by a monitored mirror reports an I/O error, +the failure is handled according to +\fBmirror_image_fault_policy\fP and \fBmirror_log_fault_policy\fP +set in \fBlvm.conf\fP. +.TP +.I \-\-ignoremonitoring +Make no attempt to interact with dmeventd unless \-\-monitor +is specified. .TP .I \-p, \-\-permission r|rw Set access permissions to read only or read and write.
man/vgchange.8.in +1 -2
@@ -78,8 +78,7 @@ are not marked as clustered. Generate new random UUID for specified Volume Groups. .TP .BR \-\-monitor " " { y | n } -Controls whether or not a mirrored logical volume is monitored by -dmeventd, if it is installed. +Start or stop monitoring a logical volume, if dmeventd is installed. If a device used by a monitored mirror reports an I/O error, the failure is handled according to .BR mirror_image_fault_policy
tools/commands.h +9 -5
@@ -150,6 +150,8 @@ xx(lvcreate,"\t[-C|--contiguous {y|n}]\n" "\t[-d|--debug]\n" "\t[-h|-?|--help]\n" +"\t[--ignoremonitoring]\n" +"\t[--monitor {y|n}]\n" "\t[-i|--stripes Stripes [-I|--stripesize StripeSize]]\n" "\t{-l|--extents LogicalExtentsNumber[%{VG|PVS|FREE}] |\n" "\t -L|--size LogicalVolumeSize[bBsSkKmMgGtTpPeE]}\n"
@@ -177,6 +179,8 @@ xx(lvcreate,"\t[-C|--contiguous {y|n}]\n" "\t[-d|--debug]\n" "\t[-h|-?|--help]\n" +"\t[--ignoremonitoring]\n" +"\t[--monitor {y|n}]\n" "\t[-i|--stripes Stripes [-I|--stripesize StripeSize]]\n" "\t{-l|--extents LogicalExtentsNumber[%{VG|FREE|ORIGIN}] |\n" "\t -L|--size LogicalVolumeSize[bBsSkKmMgGtTpPeE]}\n"
@@ -192,11 +196,11 @@ xx(lvcreate, "\t[PhysicalVolumePath...]\n\n", addtag_ARG, alloc_ARG, autobackup_ARG, chunksize_ARG, contiguous_ARG, - corelog_ARG, extents_ARG, major_ARG, minor_ARG, mirrorlog_ARG, mirrors_ARG, - name_ARG, nosync_ARG, noudevsync_ARG, permission_ARG, persistent_ARG, - readahead_ARG, regionsize_ARG, size_ARG, snapshot_ARG, stripes_ARG, - stripesize_ARG, test_ARG, type_ARG, virtualoriginsize_ARG, virtualsize_ARG, - zero_ARG) + corelog_ARG, extents_ARG, ignoremonitoring_ARG, major_ARG, minor_ARG, + mirrorlog_ARG, mirrors_ARG, monitor_ARG, name_ARG, nosync_ARG, noudevsync_ARG, + permission_ARG, persistent_ARG, readahead_ARG, regionsize_ARG, size_ARG, + snapshot_ARG, stripes_ARG, stripesize_ARG, test_ARG, type_ARG, + virtualoriginsize_ARG, virtualsize_ARG, zero_ARG) xx(lvdisplay, "Display information about a logical volume",
tools/lvchange.c +5 -4
@@ -518,7 +518,7 @@ static int lvchange_single(struct cmd_context *cmd, struct logical_volume *lv, void *handle __attribute((unused))) {int doit =0 , docmds =0 ; -int archived =0 ; +int dmeventd_mode, archived =0 ;struct logical_volume *origin; if (!(lv->vg->status & LVM_WRITE) &&
@@ -575,9 +575,10 @@ static int lvchange_single(struct cmd_context *cmd, struct logical_volume *lv,return ECMD_FAILED ; } -init_dmeventd_monitor (arg_int_value (cmd, monitor_ARG, - (is_static () ||arg_count (cmd, ignoremonitoring_ARG)) ? -DMEVENTD_MONITOR_IGNORE :DEFAULT_DMEVENTD_MONITOR )); +if (!get_activation_monitoring_mode (cmd, &dmeventd_mode)) +return ECMD_FAILED ; + +init_dmeventd_monitor (dmeventd_mode);/* * FIXME: DEFAULT_BACKGROUND_POLLING should be "unspecified".
tools/lvcreate.c +3 -0
@@ -482,6 +482,9 @@ static int _lvcreate_params(struct lvcreate_params *lp,return 0 ; } +if (!get_activation_monitoring_mode (cmd, &lp->activation_monitoring)) + return_0; + if (!_lvcreate_name_params(lp, cmd, &argc, &argv) || !_read_size_params(lp, lcp, cmd) || !_read_stripe_params(lp, cmd) ||
tools/toollib.c +21 -0
@@ -1423,3 +1423,24 @@ int pvcreate_params_validate(struct cmd_context *cmd,return 1 ; } +int get_activation_monitoring_mode (struct cmd_context *cmd, +int *monitoring_mode) +{ + *monitoring_mode =DEFAULT_DMEVENTD_MONITOR ; + +if (arg_count (cmd, monitor_ARG) && +arg_count (cmd, ignoremonitoring_ARG)) { +log_error ("Conflicting monitor and ignoremonitoring options" ); +return 0 ; + } + +if (arg_count (cmd, monitor_ARG)) + *monitoring_mode =arg_int_value (cmd, monitor_ARG, +DEFAULT_DMEVENTD_MONITOR ); +else if (is_static () ||arg_count (cmd, ignoremonitoring_ARG) || + !find_config_tree_bool (cmd,"activation/monitoring" , +DEFAULT_DMEVENTD_MONITOR )) + *monitoring_mode =DMEVENTD_MONITOR_IGNORE ; + +return 1 ; +}
tools/toollib.h +3 -0
@@ -112,4 +112,7 @@ int pvcreate_params_validate(struct cmd_context *cmd, int argc, char **argv, struct pvcreate_params *pp); +int get_activation_monitoring_mode (struct cmd_context *cmd, +int *monitoring_mode); +#endif
tools/vgchange.c +5 -4
@@ -522,16 +522,17 @@ static int vgchange_single(struct cmd_context *cmd, const char *vg_name,struct volume_group *vg, void *handle __attribute((unused))) { -int r =ECMD_FAILED ; +int dmeventd_mode, r =ECMD_FAILED ;if (vg_is_exported (vg)) {log_error ("Volume group \"%s\" is exported" , vg_name);return ECMD_FAILED ; } -init_dmeventd_monitor (arg_int_value (cmd, monitor_ARG, - (is_static () ||arg_count (cmd, ignoremonitoring_ARG)) ? -DMEVENTD_MONITOR_IGNORE :DEFAULT_DMEVENTD_MONITOR )); +if (!get_activation_monitoring_mode (cmd, &dmeventd_mode)) +return ECMD_FAILED ; + +init_dmeventd_monitor (dmeventd_mode);/* * FIXME: DEFAULT_BACKGROUND_POLLING should be "unspecified".
PATCH v2 1/1 improve activation monitoring option processing
Quoted 3 lines · Mike Snitzer
Add "monitoring" option to "activation" section of lvm.conf
Have clvmd consult the lvm.conf "activation/monitoring" too.
I am not sure if this is correct, there is flag sent from calling process. I think the process should not send this flag in the first place and not read lvm.conf later.
and specifically NOT in do_lock_lv for every LV.
Quoted 9 lines
Introduce toollib.c:get_activation_monitoring_mode().
Error out when both --monitor and --ignoremonitoring are provided.
Add --monitor and --ignoremonitoring support to lvcreate. Update lvcreate man page accordingly.
Clarify that '--monitor' controls the start and stop of monitoring in the {vg,lv}change man pages.
Sorry if it is dumb question, but if user disables monitoring, do we still emit "block_on_error" for mirror table?
If so, is it correct? When a fail happens mirror all IOs waits for userspace to do something - but there is no monitoring active.
Probably ok if command run manually (like in itnitrd, because we later start dmeventd in initscripts) but if it is disabled permanently in lvm.conf?
Milan
PATCH v2 1/1 improve activation monitoring option processing
In reply to Mike SnitzerQuoted 3 lines · Mike Snitzer
Add "monitoring" option to "activation" section of lvm.conf
Have clvmd consult the lvm.conf "activation/monitoring" too.
please do it something like this... (untested)
Milan
---
daemons/clvmd/lvm-functions.c | 10 +++++----- 1 files changed, 5 insertions(+), 5 deletions(-)
daemons/clvmd/lvm-functions.c +5 -5
@@ -483,6 +483,7 @@ const char *do_lock_query(char *resource)int do_lock_lv (unsigned char command,unsigned char lock_flags,char *resource) {int status =0 ; +int dmeventd_mode =dmeventd_monitor_mode ();DEBUGLOG ("do_lock_lv: resource '%s', cmd = %s, flags = %s, memlock = %d\n" , resource,decode_locking_cmd (command),decode_flags (lock_flags),memlock ());
@@ -542,12 +543,8 @@ int do_lock_lv(unsigned char command, unsigned char lock_flags, char *resource)if (lock_flags &LCK_MIRROR_NOSYNC_MODE )init_mirror_in_sync (0 ); -if (!(lock_flags &LCK_DMEVENTD_MONITOR_MODE )) { -int dmeventd_mode = -find_config_tree_bool (cmd,"activation/monitoring" , -DEFAULT_DMEVENTD_MONITOR ); +if (!(lock_flags &LCK_DMEVENTD_MONITOR_MODE ))init_dmeventd_monitor (dmeventd_mode); - } cmd->partial_activation =0 ;
@@ -638,6 +635,9 @@ int do_refresh_cache()return -1 ; } +init_dmeventd_monitor (find_config_tree_bool (cmd, +"activation/monitoring" , +DEFAULT_DMEVENTD_MONITOR ));init_full_scan_done (0 );init_ignore_suspended_devices (1 );lvmcache_label_scan (cmd,2 );
PATCH v2 1/1 improve activation monitoring option processing
Quoted 7 lines · Milan Broz
Quoted 3 lines · Mike Snitzer
Add "monitoring" option to "activation" section of lvm.conf
Have clvmd consult the lvm.conf "activation/monitoring" too.
please do it something like this... (untested)
OK, I already committed that other patch (got ack to do so).
Your patch looks good. I had some doubts about checking lvm.conf like I did in do_lock_query (but didn't explicitly share as much).
Mike
PATCH v2 1/1 improve activation monitoring option processing
Quoted 8 lines · Mike Snitzer
Quoted 5 lines · Milan Broz
Quoted 2 lines · Mike Snitzer
Add "monitoring" option to "activation" section of lvm.conf
Quoted 7 lines
Quoted 2 lines
Quoted 1 line
Have clvmd consult the lvm.conf "activation/monitoring" too.
please do it something like this... (untested)
OK, I already committed that other patch (got ack to do so).
Your patch looks good. I had some doubts about checking lvm.conf like I did in
do_lock_query(but didn't explicitly share as much).
Update _process_config() to establish '_dmeventd_monitor' global based on lvm.conf. This allows clvmd's calls to create_toolcontext() and refresh_toolcontext() to (re)establish dmeventd monitoring based on lvm.conf (rather than always using DEFAULT_DMEVENTD_MONITOR).
clvmd's do_lock_lv() already properly controls dmeventd monitoring based on LCK_DMEVENTD_MONITOR_MODE in lock_flags -- which currently gets set based on the '_dmeventd_monitor' global.
Follow-on work will tie command line overrides into the command context to allow clvmd's dmeventd monitoring to be controlled from the commandline. So clvmd will consult the command context for dmeventd monitoring mode rather than looking at the global (which is process specific).
---
daemons/clvmd/lvm-functions.c +2 -6
@@ -542,12 +542,8 @@ int do_lock_lv(unsigned char command, unsigned char lock_flags, char *resource)if (lock_flags &LCK_MIRROR_NOSYNC_MODE )init_mirror_in_sync (0 ); -if (!(lock_flags &LCK_DMEVENTD_MONITOR_MODE )) { -int dmeventd_mode = -find_config_tree_bool (cmd,"activation/monitoring" , -DEFAULT_DMEVENTD_MONITOR ); -init_dmeventd_monitor (dmeventd_mode); - } +if (!(lock_flags &LCK_DMEVENTD_MONITOR_MODE )) +init_dmeventd_monitor (1 ); cmd->partial_activation =0 ;
lib/commands/toolcontext.c +4 -0
@@ -281,6 +281,10 @@ static int _process_config(struct cmd_context *cmd)"activation/missing_stripe_filler" , DEFAULT_STRIPE_FILLER); +init_dmeventd_monitor (find_config_tree_bool (cmd, +"activation/monitoring" , +DEFAULT_DMEVENTD_MONITOR )); +/* FIXME Missing error code checks from the stats, not log_warn?, notify if setting overridden, delay message/check till it is actually used (eg consider if lvm shell - file could appear later after this check)? */ if (!strcmp (cmd->stripe_filler,"/dev/ioerror" ) &&stat (cmd->stripe_filler, &st))
PATCH v2 1/1 improve activation monitoring option processing
Quoted 9 lines · Mike Snitzer
Update
_process_config()to establish '_dmeventd_monitor' global based on lvm.conf. This allows clvmd's calls tocreate_toolcontext()andrefresh_toolcontext()to (re)establish dmeventd monitoring based on lvm.conf (rather than always usingDEFAULT_DMEVENTD_MONITOR).clvmd's
do_lock_lv()already properly controls dmeventd monitoring based onLCK_DMEVENTD_MONITOR_MODEinlock_flags-- which currently gets set based on the '_dmeventd_monitor' global.
Actually, it looks like we never offered/tested the ability to disable monitoring across the cluster.
clvmd's do_lock_lv() will _not_ call init_dmeventd_monitor(0) unless dmeventd monitoring is disabled (which will currently never happen because '_dmeventd_monitor' defaults to enabled; the _process_config patch, described in 1st paragraph above, will address that).
Also, dmeventd_monitor_mode() does not return a binary value. That is the low hanging fruit (I'm working on a patch to appropriately replace dmeventd_monitor_mode() calls expecting a binary return with is_dmeventd_monitor_enabled()).
I think those 2 patches fix the ability to disable monitoring with clvmd. Tracing seems to show it works. Does this sound reasonable?
I'll post the patches shortly.
Mike
p.s. overview of current cluster dmeventd_monitor propagation:
static int _dmeventd_monitor = DEFAULT_DMEVENTD_MONITOR;
int dmeventd_monitor_mode(void)
{
return _dmeventd_monitor;
}lib/locking/cluster_locking.c:_lock_for_cluster() {
...
if (dmeventd_monitor_mode())
args[1] |= LCK_DMEVENTD_MONITOR_MODE;
}daemons/clvmd/lvm-functions.c:do_lock_lv() {
...
if (!(lock_flags & LCK_DMEVENTD_MONITOR_MODE))
init_dmeventd_monitor(0);
...
if (!(lock_flags & LCK_DMEVENTD_MONITOR_MODE))
init_dmeventd_monitor(DEFAULT_DMEVENTD_MONITOR);
}PATCH 1/1 allow disabled dmeventd monitoring to propagate via clvmd
The lvm commands that control dmeventd monitoring (lvchange, vgchange, lvcreate) now properly establish the '_dmeventd_monitor' via commandline of lvm.conf -- they in turn pass their setting on to clvmd through _lock_for_cluster().
So we really don't have a need to init_dmeventd_monitor() based on lvm.conf when creating or refreshing the toolcontext -- especially not for clvmd's benefit.
And all existing dmeventd_monitor_mode() callers properly account for it returning tri-state (except for _lock_for_cluster() -- it treating it as a binary return didn't dawn on me until I wasted too much time tracing this issue).
I've tested this minimalist patch quite a bit:
clvmd's do_lock_lv() already properly controls dmeventd monitoring based on LCK_DMEVENTD_MONITOR_MODE in lock_flags -- though one small fix was needed for this to work: _lock_for_cluster() must treat dmeventd_monitor_mode()'s return as a tri-state value.
---
lib/locking/cluster_locking.c +3 -1
@@ -307,6 +307,7 @@ static int _lock_for_cluster(struct cmd_context *cmd, unsigned char clvmd_cmd,char *args;const char *node ="" ;int len; +int dmeventd_mode;int saved_errno = errno;lvm_response_t *response =NULL ;int num_responses;
@@ -324,7 +325,8 @@ static int _lock_for_cluster(struct cmd_context *cmd, unsigned char clvmd_cmd,if (mirror_in_sync ()) args[1 ] |=LCK_MIRROR_NOSYNC_MODE ; -if (dmeventd_monitor_mode ()) + dmeventd_mode =dmeventd_monitor_mode (); +if (dmeventd_mode !=DMEVENTD_MONITOR_IGNORE && dmeventd_mode) args[1 ] |=LCK_DMEVENTD_MONITOR_MODE ;if (cmd->partial_activation)