System Setup

LNet Resiliency/Health is mainly implemented at the LNet layer. The LND layer is only responsible for propagating specific errors up to the LNet Layer, which then reacts to those errors as defined in the Requirements and HLD documentation.

In order to properly test this feature fine grained control over the LND behavior is required. The drop/delay message policies will be modified to simulate the various errors that could occur when sending a message. This is described in the sections below.

Selection Algorithm Scenarios

Test #	Tag	Procedure
1	SRC_SPEC_LOCAL_MR_DST	MR Node MR Peer Send a ping REPLY for PING should always come on the same interface that PING was sent on. Check the TRACE in the logs to verify Repeat the test. A different local NI should be used for each new PING.
2	SRC_SPEC_LOCAL_MR_DST	MR Node MR Peer Initiate discovery Node → PING → Peer Node ← PUSH ← Peer Node should respond with an ACK to the same interface as the one it received the PUSH on Check the TRACE in the logs to verify Repeat the test Peer's local_ni when sending the PUSH should be different.
3	SRC_SPEC_ROUTER_MR_DST	MR Node NMR Router MR Peer Send a ping REPLY for PING should always come on the same interface that PING was sent on. Check the TRACE in the logs to verify Router should be used Repeat the test. A different local NI should be used for each new PING.
4	SRC_SPEC_ROUTER_MR_DST	MR Node NMR Router MR Peer Initiate discovery Node → PING → Peer Node ← PUSH ← Peer Node should respond with an ACK to the same interface as the one it received the PUSH on Check the TRACE in the logs to verify Router should be used Repeat the test. Peer's local_ni when sending the PUSH should be different.
5	SRC_SPEC_ROUTER_MR_DST	MR Node MR Router MR Peer Send a ping REPLY for PING should always come on the same interface that PING was sent on. Check the TRACE in the logs to verify Repeat sending Router interfaces should be used in round robin, while the peer destination should remain constant. Repeat the test. A different local NI should be used for each new PING.
6	SRC_SPEC_ROUTER_MR_DST	MR Node MR Router MR Peer Initiate discovery Node → PING → Peer Node ← PUSH ← Peer Node should respond with an ACK to the same interface as the one it received the PUSH on Check the TRACE in the logs to verify Router interfaces should be used in round robin, while the peer destination should remain constant. Repeat the test. Peer's local_ni when sending the PUSH should be different.
7	SRC_SPEC_LOCAL_NMR_DST	Same as 1 and 2 Except that repeating the test will not result in a different local_ni being used.
8	SRC_SPEC_ROUTER_NMR_DST	Same as 3 - 6 Except that repeating the test will not result in different local NIs being used.
9	SRC_ANY_LOCAL_MR_DST	MR Node MR Peer Send multiple PINGs PING REPLYs should come on the same interface Every PING will select a new local/remote NIs
10	SRC_ANY_ROUTER_MR_DST	MR Node NMR Router MR Peer Send Multiple PINGs Node will cycle over local_NIs Node will use the same destination NID as final destination Node will use the NMR Router
11	SRC_ANY_ROUTER_MR_DST	MR Node MR Router MR Peer Send Multiple PINGs Node will cycle over local_NIs Node will use the same destination NID as final destination Node will use the different interfaces of the MR Router MR Router will cycle over the interfaces of the Final destination.
12	SRC_ANY_LOCAL_NMR_DST	MR Node NMR Peer Send multiple PINGs Node will use same source/dst NID for all PINGs
13	SRC_ANY_ROUTER_NMR_DST	MR Node NMR Router NMR Peer Send multiple PINGs Node will use the same source/dst NIDs for all PINGs Node will use the router interface
14	SRC_ANY_ROUTER_NMR_DST	MR Node MR Router NMR Peer Send multiple PINGs Node will use the same source/dst NIDs for all PINGs Node will cycle through the Router's interfaces

Error Scenarios

Synchronous Errors

Test #	Tag	Procedure	Script	Result
1	Immediate Failure	Send a PING simulate an immediate LND failure (EX: NOMEM) Message should not be resent

Asynchronous Errors

Test #	Tag	Procedure
1	LNET_MSG_STATUS_LOCAL_INTERRUPT LNET_MSG_STATUS_LOCAL_DROPPED LNET_MSG_STATUS_LOCAL_ABORTED LNET_MSG_STATUS_LOCAL_NO_ROUTE LNET_MSG_STATUS_LOCAL_TIMEOUT	MR Node with Multiple interfaces Send a PING Simulate an <error> PING msg should be queued on resend queue PING msg will be resent on a different interface Failed interfaces' health value will be decremented Failed interface will be placed on the recovery queue
2	Sensitivity == 0	Same setup as 1 NI is not placed on the recovery queue
3	Sensitivity > 0	Same setup as 1 NI is placed on the recovery queue Monitor network activity as NI is pinged until health is back to maximum
4	Sensitivity > 0 Buggy interface	Same setup as 1 NI is placed on recovery queue NI is pinged ever 1 second Simulate ping failure ever other ping NI's health should be decremented on failure NI should remain on the recovery queue
5	Retry count == 0	Same setup as 1 Message will not be retried and the message will be finalized immediately
6	Retry count > 0	Same setup as 1 Message will be transmitted for a maximum of retry count or until the message expires
7	REPLY timeout	Same setup as 1 Except Use LNet selftest Simulate a local timeout Re-transmit No REPLY received Message is finalized and TIMEOUT event is propagated.
8	ACK timeout	Same setup as 7 except simulate ACK timeout
9	LNET_MSG_STATUS_LOCAL_ERROR	Same setup as 1 Message is finalized immediately (not resent) Local NI is placed on the recovery queue Same procedure to recover the local NI
10	LNET_MSG_STATUS_REMOTE_DROPPED	Same setup as 1 Message is queued for resend depending on retry_count peer_ni is placed on the recovery queue (not if sensitivity == 0) peer_ni is pinged every 1 second
11	LNET_MSG_STATUS_REMOTE_ERROR LNET_MSG_STATUS_REMOTE_TIMEOUT LNET_MSG_STATUS_NETWORK_TIMEOUT	Same setup as 1 Message is not resent peer_ni recovery happens as outlined in previous cases

Random Failures

Test #	Tag	Procedure
1	self test	MR Node NMR Peer Self-test Randomize local NI failure Randomize Remote NI failure
2	self test	MR Node MR Peer Self-test Randomize local NI failure Randomize Remote NI failure
3	self test	MR Node MR Router NMR Peer Self-test Randomize local NI failure Randomize Remote NI failure
4	self test	MR Node MR Router MR Peer Self-test Randomize local NI failure Randomize Remote NI failure
5	self test	MR Node NMR Router NMR Peer Self-test Randomize local NI failure Randomize Remote NI failure
6	self test	MR Node NMR Router MR Peer Self-test Randomize local NI failure Randomize Remote NI failure

MR Router Testing

Test #	Tag	Procedure	Script	Result
	Discovery triggered on route add	Bring up Router A with two interfaces tcp0 tcp1 Bring up Peer A and add network on tcp0 Add router to tcp1 on peerA Observe that a discovery occurs from peer A→ Router A
	Discovery triggered on interval	Bring up Router A with two interfaces tcp0 tcp1 Bring up Peer A and add network on tcp0 Add router to tcp1 on peerA Observe that a discovery occurs from peer A→ Router A Keep the two nodes up for 4 minutes Every router_interval_timeout a discovery should occur from peerA→ RouterA
	Router tcp1 down due to no traffic	Bring up Router A with two interfaces tcp0 tcp1 Bring up Peer A and add network on tcp0 Add router to tcp1 on peerA Observe that a discovery occurs from peer A→ Router A Keep the two nodes up for 4 minutes Every router_interval_timeout a discovery should occur from peerA→ RouterA Since there is no traffic on tcp1 RouterA tcp1 should be down verify via: lnetctl net show -v
	Router tcp1 comes up when peerB is brought up	Bring up Router A with two interfaces tcp0 tcp1 Bring up Peer A and add network on tcp0 Add router to tcp1 on peerA Observe that a discovery occurs from peer A→ Router A Keep the two nodes up for 4 minutes Every router_interval_timeout a discovery should occur from peerA→ RouterA Since there is no traffic on tcp1 RouterA tcp1 should be down verify via: lnetctl net show -v Bring up Peer B and add network on tcp1 Add router to tcp on peer B Observe that a discovery occurs from peerB → RouterA Observe that a RouterA tcp1 is now up
	Add route without router there	Bring up Peer A and add network on tcp0 Add route to tcp1 on peerA Observe that a discovery occurs but no response since router is not up lnetctl route show -v # shows that router is down lnetctl peer show -v # shows the peer is down Bring up Router A with two interfaces: tcp0, tcp1 After router_interval_timeout a discovery should verify that router A is up lnetctl route show -v # shows that router is down because no routerA tcp1 network should be down lnetctl peer show -v # shows the peer is up Bring up PeerB and add network on tcp1 lnetctl route show -v # shows that router is up
	traffic should trigger an attempt at router discovery	Bring up Peer A and add network on tcp0 Add route to tcp1 on peerA Observe that a discovery occurs but no response since router is not up lnetctl route show -v # shows that router is down lnetctl peer show -v # shows the router is down Bring up Router A with two interfaces: tcp0, tcp1 Bring up PeerB and add network on tcp1 Before the router_interval_timeout expires do a: lnetctl discover PeerB@tcp1 This should trigger a discovery of router A lnetctl peer show -v # shows the peer is up and multi-rail lnetctl route show -v # shows the route up
	Ping should not trigger discovery of router	Bring up Peer A and add network on tcp0 Add router to tcp1 on peerA Observe that a discovery occurs but no response since router is not up lnetctl route show -v # shows that router is down lnetctl peer show -v # shows the router is down Bring up Router A with two interfaces: tcp0, tcp1 Bring up PeerB and add network on tcp1 Before the router_interval_timeout expires do a: lnetctl ping PeerB@tcp1 This should NOT trigger a discovery of router A ping should fail lnetctl peer show -v # shows the peer is down lnetctl route show -v # shows the route down
	Multi-interface router even traffic distribution	Bring up Router A with 4 interfaces. 2 on tcp0 and 2 on tcp1 Bring up Peer A with interface on tcp0 Bring up Peer B with interface on tcp1 Run traffic using selftest Observe that traffic is distributed on all router interfaces evenly
	Multi-interface router with one bad interface	Bring up Router A with 4 interfaces. 2 on tcp0 and 2 on tcp1 Bring up Peer A with interface on tcp0 Bring up Peer B with interface on tcp1 Run traffic using selftest Observe that traffic is distributed on all router interfaces evenly Enable health (sensitivity, retries) Add a PUT drop rule on the router to drop traffic on one of the interfaces in tcp0 Observe that traffic goes to the other interfaces. There shouldn't be any drop in traffic. As long as the interface has less than optimal health, it should never be used for routing.
	Multi-interface router with a bad interface that recovers	Bring up Router A with 4 interfaces. 2 on tcp0 and 2 on tcp1 Bring up Peer A with interface on tcp0 Bring up Peer B with interface on tcp1 Run traffic using selftest Observe that traffic is distributed on all router interfaces evenly Enable health (sensitivity, retries) Add a PUT drop rule on the router to drop traffic on one of the interfaces in tcp0 Observe that traffic goes to the other interfaces. There shouldn't be any drop in traffic. As long as the interface has less than optimal health, it should never be used for routing. Remove the PUT drop rule from the router Eventually that interface should be healthy again Traffic should resume using that interface
	Multi-Router/Multi-interface setup	Bring up router A with 4 interfaces 2 on tcp0 and 2 on tcp1 Bring up router B with 4 interfaces 2 on tcp0 and 2 on tcp1 Bring up Peer A with interface on tcp0 Bring up Peer B with interface on tcp1 Run traffic Observe that traffic is distributed evenly on the interfaces of router A and B
	Multi-Router/Multi-interface setup with failed gateway	Bring up router A with 4 interfaces 2 on tcp0 and 2 on tcp1 Bring up router B with 4 interfaces 2 on tcp0 and 2 on tcp1 Bring up Peer A with interface on tcp0 Bring up Peer B with interface on tcp1 Run traffic Observe that traffic is distributed evenly on the interfaces of router A and B Add a drop rule on Router A that impacts all of its interfaces Observe that traffic is diverted to Router B with no drop in traffic.
	Multi-Router/Multi-interface setup with router recovery	Bring up router A with 4 interfaces 2 on tcp0 and 2 on tcp1 Bring up router B with 4 interfaces 2 on tcp0 and 2 on tcp1 Bring up Peer A with interface on tcp0 Bring up Peer B with interface on tcp1 Run traffic Observe that traffic is distributed evenly on the interfaces of router A and B Add a drop rule on Router A that impacts all of its interfaces Observe that traffic is diverted to Router B with no failure. Remove the rule from Router A Observe that traffic starts going through Router A again. There should be no drop in traffic
	Extra Health Testing	Run through the health test cases above while there exists a multi-rail router.

User Interface

Test #	Tag	Procedure
1	lnet_transaction_timeout	Set lnet_transaction_timeout to a value < retry_count via lnetctl and YAML This should lead to a failure to set Set lnet_transaction_timeout to a value > retr_count via lnetctl and YAML lnet_lnd_timeout value should == lnet_transaction_timeout / retry_count Show value via "lnetctl global show"
2	lnet_retry_count	Set the lnet_retry_count to a value > lnet_transaction_timeout via lnetctl and YAML This should lead to a failure to set Set the lnet_retry_count to a value < lnet_transaction_timeout via lnetctl and YAML lnet_lnd_timeout value should == lnet_transaction_timeout / retry_count Show value via "lnetctl global show"
3	lnet_health_sensitivity	Set the lnet_health sensitivity from lnetctl and from YAML Show value via "lnetctl global show"
4	NI statistics	verify LNet health statistics lnetctl net show -v 3
5	Peer NI statistics	verify LNet health statistics for peer NIs lnetctl peer show -v 3
6	NI Health value	verify setting the local NI health statistics lnetctl net set --nid <nid> --health <value> Redo from YAML
7	Peer NI Health value	verify setting the local NI health statistics lnetctl peer set --nid <nid> --health <value> Redo from YAML

Testing Tools

The drop policy has been modified to drop outgoing messages with specific errors. This can be done via the following commands. Unfortunately, for details on these commands you'll need to look at the code. A combination of these commands on the different nodes should cover approximately 75% of the health code paths.

lctl net_drop_add -s *@tcp -d *@tcp -m ACK -i 20
lctl net_drop_add -s *@tcp -d *@tcp -m REPLY -i 20
lctl net_drop_add -s *@tcp -d *@tcp -m GET -i 43 -e random -n
lctl net_drop_add -s *@tcp -d *@tcp -m PUT -i 20 -e random
lctl net_drop_del -s *@tcp -d *@tcp

The -e parameter can take the following arguments

local_interrupt 		# will result in a resend
local_dropped			# will result in a resend
local_aborted			# will result in a resend
local_no_route			# will result in a resend
local_error				# will not result in a resend
local_timeout			# will result in a resend
remote_error			# will not result in a resend
remote_dropped			# will result in a resend
remote_timeout			# will not result in a resend
network_timeout			# will not result in a resend
random
silent_queue			# will queue the message and never call lnet_finalize()

The -e can be repeated multiple times to specify a set of different errors to select randomly from. random can be given to -e to select any error to simulate at random.

Simulation Details

The error simulation occurs immediately before putting the message on the wire. If the drop rule policy is defined and is matched, then the message is not sent and error is simulated.

Types of Error Simulation Testing

Type

Description

Drop with error

This is the newly added error simulation. And it is designed to simulate different health failures. This can be used to exercise the following scenarios

re-transmit message scenair
Immediate failure and message finalization
Local NI recovery
Peer NI recovery
Finalizing active message

Drop Received messages

This an existing rule and it can be used to drop received GET/PUT messages. This will result in no ACK/REPLY being sent to the message initiator and will exercise the response timeout code.

Queuing messages

In a heavily used systems, especially routers, the credits can dip below zero and a message can be queued internally. It's possible that these message can be queued for a long period of time, so a mechanims was created to finalize these messages after the expiration time. A command can be issued to simulate queueing. This will queue the message on a separate queue which is never checked except when the message expires. The message credit is not returned until the message expires.

Configuration

LNet Health is off by default. To turn it on, two configuration parameters need to be set

Retry counter. This will indicate how many times a message should be resent before it succeeds or times out. It default to 0, which means no message re-transmission will occur.
1. lnetctl set retry_count <value>
Health sensitivity. This is a value by which to decrement the health of the interface. When an interface health value goes below the optimal value, it gets placed on a recovery queue and will be pinged every second until its health recovers. This value by default is 0, which means that a peer or local NI will never go into recovery state.
1. lnetctl set health_sensitivity <value>
Transaction timeout. This is the timeout value to wait before a response expires or before a message on the active list expires.
1. lnetctl set transaction_timeout <value>

Space shortcuts

Page tree

System Setup

Selection Algorithm Scenarios

Error Scenarios

Synchronous Errors

Asynchronous Errors

Random Failures

MR Router Testing

User Interface

Testing Tools

Simulation Details

Types of Error Simulation Testing

Configuration

Space shortcuts

Page tree

LNet Health Test Plan

System Setup

Selection Algorithm Scenarios

Error Scenarios

Synchronous Errors

Asynchronous Errors

Random Failures

MR Router Testing

User Interface

Testing Tools

Simulation Details

Types of Error Simulation Testing

Configuration