cfsd.conf(4)

Index for
Section 4

Alphabetical
listing for C

Bottom of
page

cfsd.conf(4)
NAME
cfsd.conf - cfsd daemon configuration file

DESCRIPTION
The /etc/cfsd.conf configuration file specifies operational attributes for
the cfsd daemon and defines sets of file system placement policies. In a
cluster, each file system has a placement policy associated with it. This
policy determines how cfsd manages the file system.

If you modify this file, keep the following points in mind:

· All file systems in the cluster have a placement policy associated
with them. If you do not explicitly assign a file system to a policy,
it is assigned to the default policy.

· There is no limit on the number of policies you can create.

/etc/cfsd.conf uses the stanza file format described in stanza(4).

cfsd Stanza

This stanza contains attributes that control the overall behavior of the
cfsd daemon. These attributes affect all file systems and members. There
can be only one cfsd entry in the configuration file. The following
attributes and values are recognized for the cfsd entry:

active = [1|0]
The active attribute specifies whether or not cfsd is active. If
you specify 0, cfsd is dormant: it does not monitor the cluster,
collect statistics, analyze statistics, or relocate file systems.
cfsd remains dormant until you set the active attribute to 1 and
you send cfsd a SIGHUP signal, which causes it to reread the
/etc/cfsd.conf file. (If you send SIGHUP to any cfsd daemon process
in the cluster, all cfsd daemons reread the file.)

When active is set to 1, cfsd is active to the extent specified in
its configuration file.

The default value is 0, not active.

analyze_interval = {hours:mins}[:{secs}]
The analyze_interval attribute specifies how often cfsd analyzes
the data that it has collected and subsequently writes the
/var/cluster/cfs/analysis.log file.

If analyze_interval is set to 0, cfsd does not perform analyses.

The default value for analyze_interval is 02:00:00; statistics are
analyzed and subsequently logged every two hours.

analyze_samplesize = {hours:mins}[:{secs}]
The analyze_samplesize attribute specifies the maximum amount of
data that cfsd should use in each analysis, if available. If the
amount of data indicated by analyze_samplesize is not available,
cfsd uses what is available. For each analysis, cfsd considers the
last hours, minutes, and, optionally, seconds worth of data.

The default value for analyze_samplesize is 24:00, which means that
up to the last 24 hours worth of data is used if available.

reloc_on_memory_warning = 0|1
The reloc_on_memory_warning attribute instructs cfsd how to react
to a CFS high memory usage warning on a member. If
reloc_on_memory_warning is set to 0, cfsd ignores the warning. If
set to 1, cfsd attempts to reduce CFS memory usage on the member by
moving some of the file systems that it serves to other members in
the cluster.

The default value for reloc_on_memory_warning is 0.

reloc_stagger = {mins}[:{secs}]
The reloc_stagger attribute specifies how far apart to stagger
automatic relocations. To prevent an unreasonable disruption in
cluster activity during automatic relocations, cfsd spaces the
relocations out by the number of minutes and seconds that you
specify. The default value for reloc_stagger is 0, which indicates
that automatic relocations are not staggered.

defaultpolicy Stanza

The default policy entry contains attributes that specify default analysis
and relocation attributes for any file system that is not listed in another
policy stanza. In addition, if a policy does not specifically define an
attribute that is defined in the default policy, the other policy inherits
the value set by the default policy stanza. parameters. Do not assign a
name to the default policy. There can only be one defaultpolicy stanza in
the configuration file.

The following attributes and values are recognized for the default policy
stanza:

polling_schedule = 0 | {s_days}, {s_hours}, {i_hours}:{i_mins} [:{i_secs}]
polling_schedule specifies how often the cfsd daemon collects
information for file systems and subsequently writes the
/var/cluster/cfs/stats.member file.

For each intance of polling_schedule, on scheduled days (s_days)
during scheduled hours (s_hours), cfsd collects statistics every
interval hours (i_hours), interval minutes (i_mins), and,
optionally, interval seconds (i_secs).

There can be as many polling_schedule attributes per policy as
necessary. For example, you might want a different polling
schedule for weekends.

Specify days using 0 to 6 for Sunday to Saturday. Separate values
with a dash (-) to indicate an inclusive range. To specify
individual days, separate the days with a comma and enclose the
entire days specification in quotes, as shown in the following
example:

polling_schedule = "0,6", 0-23, 01:00:00

Specify hours using 0 to 24. Separate values with a dash (-) to
indicate an inclusive range. To specify individual hours, separate
the hours with a comma and enclose the entire hours specification
in quotes, as shown in the following example:

polling_schedule = 0-6, "0,23", 01:00:00

If you specify 0, statistics collection is not performed for the
file systems.

The default value for polling_schedule is 0-6, 0-23, 00:15:00,
which equates to any day, any hour, every 15 minutes.

placement = favored
The placement attribute controls how cfsd selects members to serve
the file systems in the policy when none of the members specified
in hosting_members attributes is available.

The only supported value is favored: if none of the members
specified in hosting_members is available, cfsd selects another
member.

hosting_members = * | {member1}[, {member2}[, ...]]
The hosting_members attribute lists each cluster member that can be
the CFS server for the file systems in this policy. When relocating
a file system or recommending relocations, cfsd assigns a file
system to one of these members. hosting_members is a restrictive
list, not a preferred list; the placement attribute controls how
the file systems are placed when none of the members specified by
hosting_members attributes is available.

The cfsd daemon treats all cluster members identified in a single
hosting_members entry equally; no ordering preference is assumed by
position in the list.

To specify an ordering of member preference, use multiple
hosting_members entries:

hosting_members = member_name
hosting_members = member_name
hosting_members = member_name
hosting_members = member_name

Within a policy, cfsd gives preference to the members listed in the
first hosting_members entry, followed by the members in the next
hosting_members entry, and so on.

If you specify an asterisk (*) as the value of a hosting_members
attribute, any member of the cluster can be the CFS server for the
file systems. Even though the asterisk refers to all cluster
members, cfsd considers device connectivity when placing the file
systems. The default value is *, all cluster members.

active_placement = 0 | {reason}[, {reason}[, ...]]
active_placement lists the events that cfsd monitors for automatic
file system relocation. (The hosting_members option determines
your preference of the members to which a file system is
relocated.)

If you specify 0 for the active_placement attribute, cfsd does not
automatically relocate the file systems in this policy under any
circumstances.

The default value for active_placement is 0, which prohibits all
automatic relocations.

You can use any of the following reasons, alone or in combination:

preference
cfsd automatically relocates a file system if a more
preferred member, as specified in the hosting_members
entry, is available. For example, if a member joining a
cluster is more preferred than is the current server, cfsd
relocates the file system to the new member.

The order in which preference and connectivity appear on
the active_placement line is important. If preference is
listed first, cfsd selects more-preferred and possibly
less-connected members over more-connected, less-preferred
nodes. If connectivity is listed first, cfsd selects
more-connected and possibly less-preferred members over
more-preferred, less-connected members.

connectivity
The cfsd daemon automatically relocates a file system to a
member that has more direct connections to the devices that
the file system uses. For example, if a member loses
connectivity to the devices used by a file system that it
serves, and if connectivity is included in the
active_placement attribute, cfsd relocates that file system
to a member that does have connectivity.

The order in which preference and connectivit appear on the
active_placement line is important. If preference is
listed first, cfsd selects more-preferred and possibly
less-connected members over more-connected, less-preferred
nodes. If connectivity is listed first, cfsd selects
more-connected and possibly less-preferred members over
more-preferred, less-connected members.

memory The cfsd daemon automatically relocates some of the file
systems from a member that is approaching its CFS memory
usage limit. CFS memory usage is limited by the
svrcfstok_max_percent kernel attribute. (See the discussion
of svrcfstok_max_percent in the Cluster Administration
guide.) As a member approaches its CFS memory usage limit,
the kernel posts an EVM event as a warning. When such an
event is posted, cfsd can attempt to free memory on that
member by relocating some of the file systems that it is
serving.

failover
If the current CFS server fails, cfsd can relocate file
systems to members identified by the hosting_members
option. Ordinarily, a file system would be placed on any
one of the members connected to the most devices for the
file system. However, if failover is included in a policy's
active_placement option, cfsd provides placement hints to
the kernel. The kernel uses these hints to place the file
systems if and when they are failed over.

The Precious and Ignore Policies

The precious and ignore policies provide useful functions but they are not
required; you can remove them, change their names, modify their content,
and so forth.

The precious policy stanza is for file systems that you do not want cfsd to
relocate. (This policy inherits the default polling_schedule.) We
recommend that cfsd not be allowed to relocate the clusterwide root, /usr,
or /var file systems. (The member boot partitions are never considered for
relocation.) We also recommend that file systems whose placements are
managed by other software, such as CAA, also be assigned to this policy.

The ignore policy stanza is for file systems for which cfsd should not
collect statistics or attempt to relocate. This policy contains all NFS
and MFS file systems that are not explicitly listed in other policies. File
Systems of these types tend to be temporary and collecting statistics for
them is usually not beneficial. CFS does not relocate any NFS or MFS file
systems.

Additional Policy Stanzas

Each policy stanza can contain member assignment and relocation policy
attributes for a set of file systems. There can be as many policy stanzas
as needed. Attributes that are not given values in a policy inherit the
values of those attributes from the default policy. If an attribute is not
specified in the default policy, it has the default value listed in this
section. The following attributes and values are recognized for each policy
stanza:

name = {policyname}
Assigns a name to this policy. It can be useful for tracking which
policy a file system is assigned when it is mounted. cfsd logs
such assignments to syslog and references the policy by this name.
The first 20 characters are recognized, additional characters are
ignored. The default value for this attribute is UNNAMED.

filesystems = {fs1}[, {fs2}[,...]]
Lists the file systems that are in this policy. You can list any
number of file systems in a filesystems line. You can specify the
file system in any of the following ways:

· The directory on which the file system is mounted; for example
/mytest/test1

· The mounted-from name; for example /dev/disk/dsk5a,
host:/mytest/test2, or my_domain#fileset

· The domain name; for example my_domain# (This syntax assigns
all filesets in the domain to this policy.)

· The file system type specified as %type; for example %advfs,
%mfs, %ufs, %nfs, %cdfs, %dvdfs

If a listed file system does not exist or is not mounted, it is
ignored until it is mounted, at which time the policy is applied.
By default, filesystems is blank.

When matching known file systems to policies, cfsd uses the
following order of precedence:

1. It first tries to match a file system with the directory on
which it is mounted.

2. If the mount-point directory is not listed in any policy, cfsd
tries to match its mount-from name, or the domain name in the
case of a domain.

3. If its mount-point directory is not listed in any policy, cfsd
tries to match its file system type.

4. If its type is not assigned to any policy, the file system is
assigned to the default policy.

For example, assume that the file system type %nfs is listed in
policy A and that the mount directory /mnt is listed in policy B.
cfsd parses this situation as follows:

· If an NFS file system is mounted at /foo, it is assigned to
policy A. If an NFS file system is mounted at /mnt, it is
assigned to policy B.

· If the same file system is listed in multiple policies, for
example, /mytest is listed in two policies, the first instance
encountered in /etc/cfsd.conf is used and the others are
ignored. A message indicating this is logged to syslog:
rye cfsd[1113713]: filesystem "/mytest" is listed in multiple policies

For AdvFS filesets, the determining factor is the policy to which
the domain is assigned; all filesets in the domain are assigned
based on the domain's policy. A domain is assigned a policy in the
following manner:

· If a domain is not explicitly assigned to a policy, then it
is assigned to a policy in which one of its filesets is
listed; for example,
filesystem = my_domain#fileset

This assignment is usually based on the most-recently-mounted
fileset from that domain that is explicitly listed in a
policy.

Note

The cdfs daemon considers an AdvFS domain to be an indivisible
entity. Relocating a file system of type AdvFS affects not only the
selected fileset, but all filesets in the same domain. The entire
domain is relocated.

There can be as many polling_schedule attributes per policy as
necessary. For example, you might want a different polling
schedule for weekends.

polling_schedule = "0,6", 0-23, 01:00:00

polling_schedule = 0-6, "0,23", 01:00:00

If you specify 0, statistics collection is not performed for the
file systems.

The default value for polling_schedule is 0-6, 0-23, 00:15:00,
which equates to any day, any hour, every 15 minutes.

placement = favored
The placement attribute controls how cfsd selects members to serve
the file systems in the policy when none of the members specified
by hosting_members attributes is available.