[Nexthop][fboss2-dev] Add fboss2-dev config arp CLI subcommands

[benoit-nexthop]

Summary

Adds a new fboss2-dev config arp <attr> <value> command family for tuning the FBOSS agent's ARP/NDP neighbor-table timers at runtime. Four attributes are supported, all can be set in a hitless config reload (no agent restart required):

• timeout -> SwitchConfig.arpTimeoutSeconds
• age-interval -> SwitchConfig.arpAgerInterval
• max-probes -> SwitchConfig.maxNeighborProbes
• stale-interval -> SwitchConfig.staleEntryInterval

Note: I was going to implement a refresh subcommand as well, to set arpRefreshSeconds. That field is declared in switch_config.thrift but is not read by ApplyThriftConfig.cpp and is absent from switch_state.thrift, so setting it would be a no-op in the agent. It is explicitly excluded from kArpValidAttrs and ignored for now.

Test Plan

Unit tests

$ ./fboss/oss/scripts/bazel.sh test \
    //fboss/cli/fboss2/test/config:cmd_config_test \
    --test_filter='CmdConfigArpTestFixture.*'

[==========] Running 10 tests from 1 test suite.
[ RUN      ] CmdConfigArpTestFixture.argValidation_valid
[       OK ] CmdConfigArpTestFixture.argValidation_valid (84 ms)
[ RUN      ] CmdConfigArpTestFixture.argValidation_badArity
[       OK ] CmdConfigArpTestFixture.argValidation_badArity (30 ms)
[ RUN      ] CmdConfigArpTestFixture.argValidation_unknownAttr
[       OK ] CmdConfigArpTestFixture.argValidation_unknownAttr (34 ms)
[ RUN      ] CmdConfigArpTestFixture.argValidation_nonInteger
[       OK ] CmdConfigArpTestFixture.argValidation_nonInteger (32 ms)
[ RUN      ] CmdConfigArpTestFixture.argValidation_negative
[       OK ] CmdConfigArpTestFixture.argValidation_negative (31 ms)
[ RUN      ] CmdConfigArpTestFixture.setTimeout
[       OK ] CmdConfigArpTestFixture.setTimeout (82 ms)
[ RUN      ] CmdConfigArpTestFixture.setAgeInterval
[       OK ] CmdConfigArpTestFixture.setAgeInterval (103 ms)
[ RUN      ] CmdConfigArpTestFixture.setMaxProbes
[       OK ] CmdConfigArpTestFixture.setMaxProbes (75 ms)
[ RUN      ] CmdConfigArpTestFixture.setStaleInterval
[       OK ] CmdConfigArpTestFixture.setStaleInterval (76 ms)
[ RUN      ] CmdConfigArpTestFixture.idempotentSameValue
[       OK ] CmdConfigArpTestFixture.idempotentSameValue (85 ms)
[----------] 10 tests from CmdConfigArpTestFixture (636 ms total)
[  PASSED  ] 10 tests.

Integration tests (on gold405 / NH-4010-F)

$ sudo /tmp/fboss2_integration_test --gtest_filter='ConfigArpTest.*'

[==========] Running 4 tests from 1 test suite.
[ RUN      ] ConfigArpTest.SetTimeout
[       OK ] ConfigArpTest.SetTimeout (173 ms)
[ RUN      ] ConfigArpTest.SetAgeInterval
[       OK ] ConfigArpTest.SetAgeInterval (108 ms)
[ RUN      ] ConfigArpTest.SetMaxProbes
[       OK ] ConfigArpTest.SetMaxProbes (141 ms)
[ RUN      ] ConfigArpTest.SetStaleInterval
[       OK ] ConfigArpTest.SetStaleInterval (108 ms)
[----------] 4 tests from ConfigArpTest (531 ms total)
[  PASSED  ] 4 tests.

Each test reads the current value via the agent's thrift getRunningConfig(), sets a new value via the CLI, commits the session (HITLESS), verifies the running config reflects the change, then restores the original value.

Sample usage

$ fboss2-dev config arp timeout 90
Set arp timeout to 90
$ fboss2-dev config arp age-interval 7
Set arp age-interval to 7
$ fboss2-dev config arp max-probes 400
Set arp max-probes to 400
$ fboss2-dev config arp stale-interval 30
Set arp stale-interval to 30

$ fboss2-dev config session diff
--- current live config
+++ session config
@@ -29,9 +29,9 @@
   "sw": {
     "acls": [],
     "aggregatePorts": [],
-    "arpAgerInterval": 5,
+    "arpAgerInterval": 7,
     "arpRefreshSeconds": 20,
-    "arpTimeoutSeconds": 60,
+    "arpTimeoutSeconds": 90,
     "clientIdToAdminDistance": {
 ...
-    "maxNeighborProbes": 300,
+    "maxNeighborProbes": 400,
 ...
-    "staleEntryInterval": 10,
+    "staleEntryInterval": 30,
 ...

$ fboss2-dev config session commit
Config session committed successfully as 280a583 and config reloaded for fboss_sw_agent.