The in.mpathd daemon performs failure and repair detection for IP interfaces that have been placed into an IPMP group (or optionally, for all IP interfaces on the system). It also controls which IP interfaces in an IPMP group are “active” (being used by the system to send or receive IP data traffic) in a manner that is consistent with the administrator's configured policy.
The in.mpathd daemon can detect IP interface failure and repair through three methods: by monitoring the IFF_PHYSRUNNING flag for each IP interface (link-state based failure detection), by sending and receiving ICMP probes on each IP interface (ICMP probe-based failure detection), or by transitive probing.
If transitive probing is enabled, IP interfaces are split into two classes: those that are eligible to receive inbound IP data traffic (see INBOUND in ipmpstat(1M)) and those that are not. Interfaces that are eligible to receive inbound IP data traffic detect failures by sending and receiving ICMP probes. Those that are not eligible to receive IP data traffic detect failures by exchanging link-layer (“transitive”) probes with interfaces that are eligible.
By default, only link-state based failure detection is enabled. This requires the driver to support link-status notification. ICMP probe-based failure detection must be enabled through the configuration of one or more test addresses, which are described below. Transitive probing can be enabled by modifying the value of the SMF property (shown below) to true:
See EXAMPLES for more information on how to modify the value of the transitive-probing property.
Both ICMP and transitive probe-based failure detection methods test the entire IP interface send and receive path. The ipmpstat(1M) utility can be used to check which failure detection methods are enabled.
If transitive-probing is set to true, and no test addresses are configured for a given IPMP group, then transitive probing will be used. If it is set to false (default value), then transitive probing will not be used under any circumstance.
If only link-state based failure detection is enabled, then the health of the interface is determined solely from the state of the IFF_PHYSRUNNING flag. If probes have been enabled, the interface is considered failed if either link-state or probes indicate a failure, and repaired once the failure detection method has indicated the failure has been corrected. Although all interfaces in a group need not be configured with the same failure detection methods, transitive probing will be disabled on any interface of a group that has at least one IP test address.
As mentioned above, to perform ICMP probe-based failure detection, in.mpathd requires a test address on each IP interface for the purpose of sending and receiving probes. Any address created on an underlying interface with ipadm(1M) is automatically used as a test address. The system will automatically set the NOFAILOVER flag on such addresses. Each address may be configured statically or acquired by means of DHCP. To find targets, in.mpathd first consults the routing table for routes on the same subnet, and uses the specified next-hop. If no routes match, it sends all-hosts ICMP probes and selects a subset of the systems that respond. Thus, for probe-based failure detection to operate, there must be at least one neighbor on each subnet that responds to ICMP echo request probes. The ipmpstat(1M) utility can be used to display both the current probe target information and the status of sent probes.
Both IPv4 and IPv6 are supported. If an IP interface is plumbed for IPv4 and an IPv4 test address is configured then in.mpathd will start sending ICMPv4 probes over that IP interface. Similarly, if an IP interface is plumbed for IPv6 and an IPv6 test address is configured, then in.mpathd will start sending ICMPv6 probes over that IP interface. However, note that in.mpathd will ignore IPv6 test addresses that are not link-local. If both IPv4 and IPv6 are plumbed, it is sufficient to configure only one of the two, that is, either an IPv4 test address or an IPv6 test address. If both IPv4 and IPv6 test addresses are configured, in.mpathd probes using both ICMPv4 and ICMPv6.
As mentioned above, in.mpathd also controls which IP interfaces in an IPMP group are “active” (used by the system to send and receive IP data traffic). Specifically, in.mpathd tracks the administrative configuration of each IPMP group and attempts to keep the number of active IP interfaces in each group consistent with that configuration. Therefore, if an active IP interface fails, in.mpathd will activate an INACTIVE interface in the group, provided one exists (it will prefer INACTIVE interfaces that are also marked STANDBY). Likewise, if an IP interface repairs and the resulting repair leaves the IPMP group with more active interfaces than the administrative configuration specifies, in.mpathd will deactivate one of the interfaces (preferably one marked STANDBY), except when the FAILBACK variable is used, as described below. Similar adjustments will be made by in.mpathd when offlining IP interfaces (for instance, in response to if_mpadm(1M)).
The in.mpathd daemon accesses three variable values in /etc/default/mpathd: FAILURE_DETECTION_TIME, FAILBACK and TRACK_INTERFACES_ONLY_WITH_GROUPS.
The FAILURE_DETECTION_TIME variable specifies the probe-based failure detection time. The shorter the failure detection time, the more probe traffic. The default value of FAILURE_DETECTION_TIME is 10 seconds. This means that IP interface failure will be detected by in.mpathd within 10 seconds. The IP interface repair detection time is always twice the value of FAILURE_DETECTION_TIME. Note that failures and repairs detected by link-based failure detection are acted on immediately, though in.mpathd may ignore link state changes if it suspects that the link state is flapping due to defective hardware; see DIAGNOSTICS.
By default, in.mpathd limits failure and repair detection to IP interfaces that are configured as part of a named IPMP group. Setting TRACK_INTERFACES_ONLY_WITH_GROUPS to no enables failure and repair detection on all IP interfaces, even if they are not part of a named IPMP group. IP interfaces that are tracked but not part of a named IPMP group are considered to be part of the “anonymous” IPMP group. In addition to having no name, this IPMP group is special in that its IP interfaces are not equivalent and thus cannot take over for one another in the event of an IP interface failure. That is, the anonymous IPMP group can only be used for failure and repair detection, and provides no high-availability or load-spreading.
As described above, when in.mpathd detects that an IP interface has repaired, it activates it so that it will again be used to send and receive IP data traffic. However, if FAILBACK is set to no, then the IP interface will only be activated if no other active IP interfaces in the group remain. However, the interface may subsequently be activated if another IP interface in the group fails.
The in.mpathd daemon service is managed by the service management facility, smf(5), under the service identifier:
Administrative actions on this service, such as enabling, disabling, or requesting restart, can be performed using svcadm(1M).
The following example shows the sequence of SMF commands used to enable fault detection by transitive probing.
# svccfg -s svc:/network/ipmp setprop config/transitive-probing=true # svcadm refresh svc:/network/ipmp:default
Contains default values used by the in.mpathd daemon.
See attributes(5) for descriptions of the following attributes:
IP interface interface_name has a hardware address which is not unique in group group_name; offlining
Description: For probe-based failure detection, load-spreading, and other code IPMP features to work properly, each IP interface in an IPMP group must have a unique hardware address. If this requirement is not met, in.mpathd will automatically offline all but one of the IP interfaces with duplicate hardware addresses.
IP interface interface_name now has a unique hardware address in group group_name; onlining
Description: The previously-detected duplicate hardware address is now unique, and therefore in.mpathd has brought interface_name back online.
Test address address is not unique in group; disabling probe-based failure detection on interface_name
Description: For in.mpathd to perform probe-based failure detection, each test address in the group must be unique.
No test address configured on interface interface_name disabling probe-based failure detection on it
Description: For in.mpathd to perform probe-based failure detection on an IP interface, it must be configured with a test address: IPv4, IPv6, or both.
IP interface_name in group group_name is not plumbed for IPv[4|6], affecting IPv[4|6] connectivity
Description: All IP interfaces in a multipathing group must be homogeneously plumbed. For example, if one IP interface is plumbed for IPv4, then all IP interfaces in the group must be plumbed for IPv4, or IPv4 packets will not be able to be reliably sent and received. The STREAMS modules pushed on all IP interfaces must also be identical.
The link has come up on interface_name more than 2 times in the last minute; disabling repair until it stabilizes.
Description: To limit the impact of interfaces with intermittent hardware (such as a bad cable), in.mpathd will not consider an IP interface with a frequently changing link state as repaired until the link state stabilizes.
Invalid failure detection time of time, assuming default 10000 ms
Description: An invalid value was encountered for FAILURE_DETECTION_TIME in the /etc/default/mpathd file.
Too small failure detection time of time, assuming minimum of 100 ms
Description: The minimum value that can be specified for FAILURE_DETECTION_TIME is currently 100 milliseconds.
Invalid value for FAILBACK value
Description: Valid values for the boolean variable FAILBACK are yes or no.
Invalid value for TRACK_INTERFACES_ONLY_WITH_GROUPS value
Description: Valid values for the boolean variable TRACK_INTERFACES_ONLY_WITH_GROUPS are yes or no.
Cannot meet requested failure detection time of time ms on (inet interface_name) new failure detection time for group group_name is time ms
Description: The round trip time for ICMP probes is higher than necessary to maintain the current failure detection time. The network is probably congested or the probe targets are loaded. in.mpathd automatically increases the failure detection time to whatever it can achieve under these conditions.
Improved failure detection time time ms on (inet interface_name) for group group_name
Description: The round trip time for ICMP probes has now decreased and in.mpathd has lowered the failure detection time correspondingly.
IP interface failure detected on interface_name
Description: in.mpathd has detected a failure on interface_name, and has set the IFF_FAILED flag on interface_name, ensuring that it will not be used for IP data traffic.
IP interface repair detected on interface_name
Description: in.mpathd has detected a repair on interface_name, and has cleared the IFF_FAILED flag. Depending on the administrative configuration, the interface_name may again be used for IP data traffic.
All IP interfaces in group group are now unusable
Description: in.mpathd has determined that none of the IP interfaces in group can be used for IP data traffic, breaking network connectivity for the group.
At least 1 IP interface (interface_name) in group group is now usable
Description: in.mpathd has determined that at least one of the IP interfaces in group can again be used for IP data traffic, restoring network connectivity for the group.
The link has gone down on interface_name
Description: in.mpathd has detected that the IFF_PHYSRUNNING flag for interface_name has been cleared, indicating that the link has gone down.
The link has come up on interface_name
Description: in.mpathd has detected that the IFF_PHYSRUNNING flag for interface_name has been set, indicating that the link has come up.