Re-sync with internal repository

The internal and external repositories are out of sync. This Pull Request attempts to brings them back in sync by patching the GitHub repository. Please carefully review this patch. You must disable ShipIt for your project in order to merge this pull request. DO NOT IMPORT this pull request. Instead, merge it directly on GitHub using the MERGE BUTTON. Re-enable ShipIt after merging.

fbshipit-source-id: 29a7b631e2673f536e3daf48fb6fa62d0651349a

Author: facebook-github-bot

README.md

@@ -11,6 +11,17 @@ are expressed as structured configurations rather than imperative scripts.
 
 See the [documentation](https://github.com/facebook/DNE-Taac) for setup instructions.
 
+## Running TAAC modules under OSS
+
+TAAC's Python modules (e.g. `taac.libs.taac_runner`) include Meta-internal imports that aren't shipped in this slice. Set `TAAC_OSS=1` so the imports take their OSS branch:
+
+```bash
+export TAAC_OSS=1
+python3 -c 'import taac.libs.taac_runner; print("ok")'
+```
+
+Some runtime functionality is stubbed in OSS mode (NDS drainer, COOP patcher, ValidationStep, AristaSSHHelper, etc.) and will raise `NotImplementedError` if invoked on a code path that requires it. Imports succeed; trying to actually use those features fails.
+
 ## License
 
 This project is licensed under the Apache License 2.0 — see the [LICENSE](LICENSE) file for details.

taac/constants.py

@@ -362,6 +362,12 @@ class BgpPlusPlusProfile(Enum):
 RDMA_PORT: str = "4791"
 DHCPV6_MULTICAST_ADDR: str = "fe80::a00:27ff:fefe:8f95"
 DHCPV6_SERVER_MULTICAST_ADDR: str = "ff02::1:2"
+# Ethernet multicast MAC for DHCPv6 servers (ff02::1:2) per RFC 2464 §7
+# (33:33: prefix + low-32-bits of IPv6 multicast addr). Used as Ethernet
+# DMAC when the IPv6 DIP is DHCPV6_SERVER_MULTICAST_ADDR; sending a
+# multicast IPv6 destination wrapped in a unicast Ethernet DMAC produces
+# a malformed frame at L2.
+DHCPV6_SERVER_MULTICAST_MAC: str = "33:33:00:01:00:02"
 DHCPV6_SERVER_PORT: str = "547"
 DHCPV6_RELAY_PORT: str = "546"
 DHCPV4_BROADCAST_ADDR: str = "255.255.255.255"

taac/custom_test_handlers/registry.py

@@ -1,18 +1,26 @@
 # Copyright (c) Meta Platforms, Inc. and affiliates.
 # pyre-unsafe
+import os
 import typing as t
 
-from taac.ai_bb.dsf.dsf_test_handler import DsfTestHandler
 from taac.custom_test_handlers.base_custom_test_handler import (
     BaseCustomTestHandler,
 )
 from taac.custom_test_handlers.patcher_cleanup_handler import (
     PatcherCleanupHandler,
 )
 
+TAAC_OSS = os.environ.get("TAAC_OSS", "").lower() in ("1", "true", "yes")
 
 # pyre-ignore
 CUSTOM_TEST_HANDLERS: t.List[BaseCustomTestHandler] = [
-    DsfTestHandler,
     PatcherCleanupHandler,
 ]
+
+if not TAAC_OSS:
+    # DsfTestHandler lives in the Meta-internal taac.ai_bb subpackage which
+    # isn't shipped in the OSS slice.
+    from taac.ai_bb.dsf.dsf_test_handler import DsfTestHandler
+
+    # pyre-ignore[6]: registry holds handler classes, annotated as instances.
+    CUSTOM_TEST_HANDLERS.append(DsfTestHandler)

taac/driver/fboss_switch.py

@@ -36,19 +36,81 @@
 # =============================================================================
 # Third-party (PyPI / OSS-compatible)
 # =============================================================================
-import bunch
+if t.TYPE_CHECKING:
+    # Keep Pyre on the real `bunch` types; the runtime fallback below would
+    # otherwise widen `bunch.Bunch` to `Bunch | SimpleNamespace`, which breaks
+    # callers (and the `import *` re-export in dne/drivers/fboss_switch.py).
+    import bunch
+else:
+    try:
+        import bunch
+    except ImportError:
+        # bunch is abandoned (last updated 2011), use types.SimpleNamespace as replacement
+        from types import SimpleNamespace as Bunch
+
+        class BunchModule:
+            """Compatibility wrapper for abandoned bunch module"""
+
+            Bunch = Bunch
+
+        bunch = BunchModule()
 
 # =============================================================================
 # FBOSS thrift types & clients (already OSS'd)
 # =============================================================================
 import neteng.fboss.bgp_thrift.types as fboss_bgp_thrift_types
 import neteng.fboss.fsdb.types as fsdb_types
 import pexpect
-from fboss.fb_thrift_clients import FbossAgentClient, FbossAgentClientWrapper
+
+TAAC_OSS = os.environ.get("TAAC_OSS", "").lower() in ("1", "true", "yes")
+
+# `t.TYPE_CHECKING or` keeps the real FbossAgentClient visible to Pyre so
+# `with self._get_fboss_agent_client() as client:` type-checks as a context
+# manager (the OSS stub below is a runtime-only fallback).
+if t.TYPE_CHECKING or not TAAC_OSS:
+    from fboss.fb_thrift_clients import FbossAgentClient, FbossAgentClientWrapper
+else:
+    # OSS stubs - fboss.fb_thrift_clients is Meta-internal
+    # Use FbossCtrl from OSS bindings as base
+    class FbossAgentClient:  # type: ignore
+        """OSS stub - using FbossCtrl from neteng.fboss.ctrl.clients"""
+
+        def __init__(self, hostname: str, port: int = 5909, timeout: int = 30):
+            from neteng.fboss.ctrl.clients import FbossCtrl
+
+            self.hostname = hostname
+            self.port = port
+            self.timeout = timeout
+            self._client = FbossCtrl
+
+    class FbossAgentClientWrapper:  # type: ignore
+        """OSS stub - context manager wrapper for FbossCtrl"""
+
+        def __init__(self, host: str, timeout: int = 30):
+            self.host = host
+            self.timeout = timeout
+            self._client = None
+
+        def __enter__(self):
+            from neteng.fboss.ctrl.clients import FbossCtrl
+
+            # In OSS, return the client class - actual instantiation happens elsewhere
+            return FbossCtrl
+
+        def __exit__(self, *args):
+            pass
+
+
 from neteng.fboss.bgp_attr.types import TBgpAfi, TIpPrefix
 from neteng.fboss.bgp_route_types.types import TBgpPath, TRibEntry
 from neteng.fboss.bgp_thrift.clients import TBgpService
-from neteng.fboss.bgp_thrift.types import TBgpPeerState, TBgpSession, TOriginatedRoute
+from neteng.fboss.bgp_thrift.types import (
+    TBgpPeerState,
+    TBgpSession,
+    TGetUpdateGroupInfoRequest,
+    TGetUpdateGroupInfoResponse,
+    TOriginatedRoute,
+)
 from neteng.fboss.ctrl.clients import FbossCtrl
 from neteng.fboss.ctrl.types import (
     AggregatePortThrift,
@@ -155,8 +217,6 @@
     to_fb_uqdn,
 )
 
-TAAC_OSS = os.environ.get("TAAC_OSS", "").lower() in ("1", "true", "yes")
-
 if not TAAC_OSS:
     from openr.py.openr.cli.utils.commands import OpenrCtrlCmd
     from openr.py.openr.clients.openr_client import get_openr_ctrl_cpp_client
@@ -1802,6 +1862,14 @@ async def async_get_bgp_sessions(self) -> Sequence[TBgpSession]:
         async with await self._get_bgp_client() as bgp_client:
             return await bgp_client.getBgpSessions()
 
+    async def async_get_update_group_info(
+        self, group_id: t.Optional[int] = None
+    ) -> TGetUpdateGroupInfoResponse:
+        """Get BGP++ Update Group info (the API behind ``show bgpcpp update-group``)."""
+        request = TGetUpdateGroupInfoRequest(group_id=group_id)
+        async with await self._get_bgp_client() as bgp_client:
+            return await bgp_client.getUpdateGroupInfo(request)
+
     @async_retryable(
         retries=30,
         sleep_time=2,
@@ -2508,11 +2576,15 @@ def _bounce_specific_interface_using_wedge_qsfp_util(
         return True
 
     async def async_do_rapid_interface_flaps(
-        self, interface_names: Tuple[str], interval_to_link_up: int, total_flaps: int
+        self,
+        interface_names: Tuple[str],
+        interval_to_link_up: int,
+        total_flaps: int,
+        down_time_sec: float = 0.1,
     ) -> None:
         for _ in range(total_flaps):
             interface_names_str = " ".join(interface_names)
-            flap_command = f"wedge_qsfp_util -tx_disable {interface_names_str} && sleep 0.1 && wedge_qsfp_util -tx_enable {interface_names_str}"
+            flap_command = f"wedge_qsfp_util -tx_disable {interface_names_str} && sleep {down_time_sec} && wedge_qsfp_util -tx_enable {interface_names_str}"
             await self.async_run_cmd_on_shell(flap_command)
             await asyncio.sleep(interval_to_link_up)
             self.logger.info("Flap command executed")
@@ -4384,12 +4456,16 @@ async def async_get_bgp_drain_state(self) -> t.Dict[str, t.Any]:
         async with await self._get_bgp_client() as bgp_client:
             drain_state = await bgp_client.getDrainState()
             return {
-                "drain_state": drain_state.drain_state.name
-                if drain_state.drain_state is not None
-                else "UNKNOWN",
-                "drained_interfaces": list(drain_state.drained_interfaces)
-                if drain_state.drained_interfaces
-                else [],
+                "drain_state": (
+                    drain_state.drain_state.name
+                    if drain_state.drain_state is not None
+                    else "UNKNOWN"
+                ),
+                "drained_interfaces": (
+                    list(drain_state.drained_interfaces)
+                    if drain_state.drained_interfaces
+                    else []
+                ),
             }
 
     async def async_get_all_interface_names(

taac/health_checks/all_health_checks.py

@@ -3,6 +3,8 @@
 import os
 import typing as t
 
+TAAC_OSS = os.environ.get("TAAC_OSS", "").lower() in ("1", "true", "yes")
+
 from taac.health_checks.abstract_health_check import (
     AbstractDeviceHealthCheck,
     AbstractIxiaHealthCheck,
@@ -42,12 +44,12 @@
 from taac.health_checks.device_health_checks.bgp_tcpdump_health_check import (
     BgpTcpdumpHealthCheck,
 )
+from taac.health_checks.device_health_checks.bgp_update_group_health_check import (
+    BgpUpdateGroupHealthCheck,
+)
 from taac.health_checks.device_health_checks.clear_counters_health_check import (
     ClearCountersHealthCheck,
 )
-from taac.health_checks.device_health_checks.cpu_utilization_health_check import (
-    CpuUtilizationHealthCheck,
-)
 from taac.health_checks.device_health_checks.device_core_dumps_health_check import (
     DeviceCoreDumpsHealthCheck,
 )
@@ -63,9 +65,6 @@
 from taac.health_checks.device_health_checks.file_exists_health_check import (
     FileExistsHealthCheck,
 )
-from taac.health_checks.device_health_checks.generic_ods_health_check import (
-    GenericOdsHealthCheck,
-)
 from taac.health_checks.device_health_checks.hardware_capacity_health_check import (
     HardwareCapacityHealthCheck,
 )
@@ -78,9 +77,6 @@
 from taac.health_checks.device_health_checks.log_parsing_health_check import (
     LogParsingHealthCheck,
 )
-from taac.health_checks.device_health_checks.memory_utilization_health_check import (
-    MemoryUtilizationHealthCheck,
-)
 from taac.health_checks.device_health_checks.oomd_kill_health_check import (
     OomdKillHealthCheck,
 )
@@ -132,15 +128,6 @@
 from taac.health_checks.device_health_checks.systemctl_active_state_health_check import (
     SystemctlActiveStateHealthCheck,
 )
-from taac.health_checks.device_health_checks.tm_reconciliation_firing_health_check import (
-    TmReconciliationFiringHealthCheck,
-)
-from taac.health_checks.device_health_checks.ucmp_traffic_distribution_health_check import (
-    UcmpTrafficDistributionHealthCheck,
-)
-from taac.health_checks.device_health_checks.unclean_exit_health_check import (
-    UncleanExitHealthCheck,
-)
 from taac.health_checks.device_health_checks.wedge_agent_configured_health_check import (
     WedgeAgentConfiguredHealthCheck,
 )
@@ -156,12 +143,6 @@
 from taac.health_checks.dsf_health_checks.dsf_fsdb_subscriber_timestamp_health_check import (
     DsfFsdbSubscriberTimestampHealthCheck,
 )
-from taac.health_checks.dsf_health_checks.dsf_pfc_health_check import (
-    DsfPfcHealthCheck,
-)
-from taac.health_checks.dsf_health_checks.dsf_traffic_rebalance_health_check import (
-    DsfTrafficRebalanceHealthCheck,
-)
 from taac.health_checks.ixia_health_checks.ixia_packet_loss_health_check import (
     IxiaPacketLossHealthCheck,
 )
@@ -209,8 +190,6 @@
 )
 from taac.health_check.health_check import types as hc_types
 
-TAAC_OSS = os.environ.get("TAAC_OSS", "").lower() in ("1", "true", "yes")
-
 HealthCheck = t.Union[
     t.Type[AbstractIxiaHealthCheck],
     t.Type[AbstractDeviceHealthCheck],
@@ -223,32 +202,33 @@
     DrainStateHealthCheck,
     DsfDrainStateHealthCheck,
     DsfFabricReachabilityHealthCheck,
-    DsfTrafficRebalanceHealthCheck,
+    # DsfTrafficRebalanceHealthCheck,  # neteng.test_infra.dne.taac dep, excluded in OSS
     DsfFsdbSessionHealthCheck,
     DsfFsdbSubscriberTimestampHealthCheck,
     NdpHealthCheck,
     IxiaPortStatsHealthCheck,
     SystemctlActiveStateHealthCheck,
     WedgeAgentConfiguredHealthCheck,
-    DsfPfcHealthCheck,
+    # DsfPfcHealthCheck,  # neteng.test_infra.dne.taac dep, excluded in OSS
     CoreDumpsHealthCheck,
     PortStateHealthCheck,
     LldpHealthCheck,
     IxiaTrafficRateHealthCheck,
     PfcWdHealthCheck,
     CpuQueueHealthCheck,
-    UncleanExitHealthCheck,
-    CpuUtilizationHealthCheck,
-    MemoryUtilizationHealthCheck,
+    # UncleanExitHealthCheck,  # ODS-dependent (taac.internal), excluded in OSS
+    # CpuUtilizationHealthCheck,  # ODS-dependent (taac.internal), excluded in OSS
+    # MemoryUtilizationHealthCheck,  # ODS-dependent (taac.internal), excluded in OSS
     BgpSessionEstablishedHealthCheck,
     BgpConvergenceHealthCheck,
     BgpGracefulRestartHealthCheck,
+    BgpUpdateGroupHealthCheck,
     HardwareCapacityHealthCheck,
     BgpStaleRouteHealthCheck,
     BgpNonBestRouteHealthCheck,
     BgpTcpdumpHealthCheck,
     L2EntryThresholdHealthCheck,
-    GenericOdsHealthCheck,
+    # GenericOdsHealthCheck,  # ODS-dependent (taac.internal), excluded in OSS
     OomdKillHealthCheck,
     EcmpGroupAndMemberCountHealthCheck,
     DeviceCoreDumpsHealthCheck,
@@ -268,7 +248,7 @@
     BgpFibProgrammingCheck,
     PortSpeedHealthCheck,
     PortSpeedSnapshotHealthCheck,
-    UcmpTrafficDistributionHealthCheck,
+    # UcmpTrafficDistributionHealthCheck,  # ODS-dependent (taac.internal), excluded in OSS
     BgpRouteCountVerificationHealthCheck,
     BgpMultipathNextHopCountHealthCheck,
     RouteConvergenceTimeHealthCheck,
@@ -282,7 +262,7 @@
     OpenrKvstoreConsistencyHealthCheck,
     AristaFbossNextHopValidityHealthCheck,
     PortChannelExpectedStateHealthCheck,
-    TmReconciliationFiringHealthCheck,
+    # TmReconciliationFiringHealthCheck,  # ODS-dependent (taac.internal), excluded in OSS
     TmKernelStateSnapshotHealthCheck,
 ]