From c270e79576dbc0de13a60431d293b347ab997e4a Mon Sep 17 00:00:00 2001
From: Michael MacDonald <mjmac@google.com>
Date: Thu, 12 Dec 2024 00:21:39 +0000
Subject: [PATCH] add new ftest skeleton

Features: telemetry
Required-githooks: true

Signed-off-by: Michael MacDonald <mjmac@google.com>
---
 src/client/dfs/io.c                           |  2 +-
 .../ftest/telemetry/dfs_client_telemetry.py   | 49 +++++++++++++++++++
 .../ftest/telemetry/dfs_client_telemetry.yaml | 46 +++++++++++++++++
 src/tests/ftest/util/telemetry_utils.py       | 30 +++++++++++-
 4 files changed, 125 insertions(+), 2 deletions(-)
 create mode 100644 src/tests/ftest/telemetry/dfs_client_telemetry.py
 create mode 100644 src/tests/ftest/telemetry/dfs_client_telemetry.yaml

diff --git a/src/client/dfs/io.c b/src/client/dfs/io.c
index 17c2ee24bb13..db8f79ab105a 100644
--- a/src/client/dfs/io.c
+++ b/src/client/dfs/io.c
@@ -308,7 +308,7 @@ dfs_writex(dfs_t *dfs, dfs_obj_t *obj, dfs_iod_t *iod, d_sg_list_t *sgl, daos_ev
 		daos_event_errno_rc(ev);
 
 	buf_size = 0;
-	if (sgl)
+	if (dfs->metrics != NULL && sgl != NULL)
 		for (i = 0; i < sgl->sg_nr; i++)
 			buf_size += sgl->sg_iovs[i].iov_len;
 
diff --git a/src/tests/ftest/telemetry/dfs_client_telemetry.py b/src/tests/ftest/telemetry/dfs_client_telemetry.py
new file mode 100644
index 000000000000..8b0cc65be56d
--- /dev/null
+++ b/src/tests/ftest/telemetry/dfs_client_telemetry.py
@@ -0,0 +1,49 @@
+"""
+  (C) Copyright 2024 Intel Corporation.
+
+  SPDX-License-Identifier: BSD-2-Clause-Patent
+"""
+from ior_utils import read_data, write_data
+from telemetry_test_base import TestWithClientTelemetry
+from telemetry_utils import CLIENT_DFS_IO_METRICS, CLIENT_DFS_OPS_METRICS
+
+
+class DFSClientTelemetry(TestWithClientTelemetry):
+    """Tests to verify DFS telemetry.
+
+    :avocado: recursive
+    """
+
+    def test_dfs_metrics(self):
+        """JIRA ID: DAOS-16837.
+
+        Verify that the DFS metrics are incrementing as expected.
+        After performing some I/O, the DFS-level metrics should look reasonable.
+
+        Test steps:
+        1) Create a pool and container
+        2) Perform some I/O with IOR
+        3) Verify that the DFS metrics are sane
+
+        :avocado: tags=all,daily_regression
+        :avocado: tags=vm
+        :avocado: tags=telemetry
+        :avocado: tags=DFSClientTelemetry,test_dfs_metrics
+        """
+        # create pool and container
+        pool = self.get_pool(connect=True)
+        container = self.get_container(pool=pool)
+
+        self.log_step('Writing data to the pool (ior)')
+        ior = write_data(self, container)
+        self.log_step('Reading data from the pool (ior)')
+        read_data(self, ior, container)
+
+        metric_names = CLIENT_DFS_OPS_METRICS + CLIENT_DFS_IO_METRICS
+
+        self.log_step('Reading dfs telemetry')
+        after_metrics = self.telemetry.collect_client_data(metric_names)
+        for metric in metric_names:
+            print(f'{metric}: {after_metrics[metric]}')
+
+        self.log_step('Test passed')
diff --git a/src/tests/ftest/telemetry/dfs_client_telemetry.yaml b/src/tests/ftest/telemetry/dfs_client_telemetry.yaml
new file mode 100644
index 000000000000..d585dc81fda4
--- /dev/null
+++ b/src/tests/ftest/telemetry/dfs_client_telemetry.yaml
@@ -0,0 +1,46 @@
+hosts:
+  test_servers: 1
+  test_clients: 1
+
+timeout: 180
+
+server_config:
+  name: daos_server
+  engines_per_host: 1
+  engines:
+    0:
+      targets: 4
+      nr_xs_helpers: 0
+      storage:
+        0:
+          class: ram
+          scm_mount: /mnt/daos
+  system_ram_reserved: 1
+
+agent_config:
+  telemetry_port: 9191
+  telemetry_retain: 30s
+  telemetry_enabled: true
+
+pool:
+  scm_size: 2G
+
+container:
+  type: POSIX
+  control_method: daos
+  dfs_oclass: SX
+
+ior: &ior_base
+  ppn: 4
+  api: DFS
+  transfer_size: 512K
+  block_size: 1M
+  dfs_oclass: SX
+
+ior_write:
+  <<: *ior_base
+  flags: "-k -v -w -W -G 1"
+
+ior_read:
+  <<: *ior_base
+  flags: "-v -r -R -G 1"
diff --git a/src/tests/ftest/util/telemetry_utils.py b/src/tests/ftest/util/telemetry_utils.py
index fb6b37d9ef31..1d11ba8401cd 100644
--- a/src/tests/ftest/util/telemetry_utils.py
+++ b/src/tests/ftest/util/telemetry_utils.py
@@ -967,6 +967,32 @@ class ClientTelemetryUtils(TelemetryUtils):
         _gen_stats_metrics("client_io_ops_tgt_update_active")
     CLIENT_IO_OPS_UPDATE_ACTIVE_METRICS = \
         _gen_stats_metrics("client_io_ops_update_active")
+    CLIENT_DFS_OPS_METRICS = [
+        "client_dfs_ops_CHMOD",
+        "client_dfs_ops_CHOWN",
+        "client_dfs_ops_CREATE",
+        "client_dfs_ops_GETSIZE",
+        "client_dfs_ops_GETXATTR",
+        "client_dfs_ops_LSXATTR",
+        "client_dfs_ops_MKDIR",
+        "client_dfs_ops_OPEN",
+        "client_dfs_ops_OPENDIR",
+        "client_dfs_ops_READ",
+        "client_dfs_ops_READDIR",
+        "client_dfs_ops_READLINK",
+        "client_dfs_ops_RENAME",
+        "client_dfs_ops_RMXATTR",
+        "client_dfs_ops_SETATTR",
+        "client_dfs_ops_SETXATTR",
+        "client_dfs_ops_STAT",
+        "client_dfs_ops_SYMLINK",
+        "client_dfs_ops_SYNC",
+        "client_dfs_ops_TRUNCATE",
+        "client_dfs_ops_UNLINK",
+        "client_dfs_ops_WRITE"]
+    CLIENT_DFS_IO_METRICS = [
+        "client_dfs_read_bytes",
+        "client_dfs_write_bytes"]
     CLIENT_IO_METRICS = CLIENT_IO_LATENCY_FETCH_METRICS +\
         CLIENT_IO_LATENCY_UPDATE_METRICS +\
         CLIENT_IO_OPS_AKEY_ENUM_ACTIVE_METRICS +\
@@ -1009,7 +1035,9 @@ class ClientTelemetryUtils(TelemetryUtils):
         CLIENT_IO_OPS_TGT_PUNCH_ACTIVE_METRICS +\
         CLIENT_IO_OPS_TGT_PUNCH_LATENCY_METRICS +\
         CLIENT_IO_OPS_TGT_UPDATE_ACTIVE_METRICS +\
-        CLIENT_IO_OPS_UPDATE_ACTIVE_METRICS
+        CLIENT_IO_OPS_UPDATE_ACTIVE_METRICS +\
+        CLIENT_DFS_OPS_METRICS +\
+        CLIENT_DFS_IO_METRICS
 
     def __init__(self, dmg, servers, clients):
         """Create a ClientTelemetryUtils object.