From 6cb9368b81a1ef95638624791f57dd0383117651 Mon Sep 17 00:00:00 2001
From: Dalton Bohning <dalton.bohning@intel.com>
Date: Tue, 3 Dec 2024 17:44:24 +0000
Subject: [PATCH] DAOS-16845 test: add dynamic pool wait_for_aggregation

Add TestPool.wait_for_aggregation to dynamically wait for pool
aggregation to complete.
Update tests to use the new function.

Test-tag: test_enospace_time_with_fg
Skip-unit-tests: true
Skip-fault-injection-test: true

Required-githooks: true

Signed-off-by: Dalton Bohning <dalton.bohning@intel.com>
---
 src/tests/ftest/nvme/enospace.py        | 24 ++++++++++++-----
 src/tests/ftest/util/test_utils_pool.py | 34 +++++++++++++++++++++++++
 2 files changed, 51 insertions(+), 7 deletions(-)

diff --git a/src/tests/ftest/nvme/enospace.py b/src/tests/ftest/nvme/enospace.py
index c7a996c110fd..9ffa2c77e696 100644
--- a/src/tests/ftest/nvme/enospace.py
+++ b/src/tests/ftest/nvme/enospace.py
@@ -464,22 +464,32 @@ def test_enospace_time_with_fg(self):
         """
         self.log.info(self.pool.pool_percentage_used())
 
-        # Enabled TIme mode for Aggregation.
+        self.log_step("Enable pool aggregation")
         self.pool.set_property("reclaim", "time")
 
+        self.log_step("Get initial pool free space")
+        initial_space = self.pool.get_pool_daos_space()
+        initial_free_scm = initial_space["s_free"][0]
+        initial_free_nvme = initial_space["s_free"][1]
+
         # Repeat the test in loop.
         for _loop in range(10):
-            self.log.info("-------enospc_time_fg Loop--------- %d", _loop)
+            self.log_step(f"Run IOR to fill the pool - enospace_time_with_fg loop {_loop}")
             self.log.info(self.pool.pool_percentage_used())
             # Run IOR to fill the pool.
             log_file = f"-loop_{_loop}".join(os.path.splitext(self.client_log))
             self.run_enospace_with_bg_job(log_file)
-            # Delete all the containers
+            self.log_step(f"Delete all containers - enospace_time_with_fg loop {_loop}")
             self.delete_all_containers()
-            # Delete container will take some time to release the space
-            time.sleep(60)
-
-        # Run last IO
+            self.log_step(f"Wait for aggregation to complete - enospace_time_with_fg loop {_loop}")
+            agg_did_complete = self.pool.wait_for_aggregation(
+                verify_scm=lambda current: current <= initial_free_scm * 1.05,
+                verify_nvme=lambda current: current <= initial_free_nvme * 1.05,
+                retries=4, interval=30)
+            if not agg_did_complete:
+                self.fail("Pool space not reclaimed after deleting all containers")
+
+        self.log_step("Run one more sanity IOR to fill 1%")
         self.start_ior_load(storage='SCM', operation="Auto_Write", percent=1)
 
     @skipForTicket("DAOS-8896")
diff --git a/src/tests/ftest/util/test_utils_pool.py b/src/tests/ftest/util/test_utils_pool.py
index f5e88d2c26cc..ee010993376e 100644
--- a/src/tests/ftest/util/test_utils_pool.py
+++ b/src/tests/ftest/util/test_utils_pool.py
@@ -1475,6 +1475,40 @@ def wait_pool_suspect_ranks(self, expected, interval=1, timeout=30):
 
         self.log.info("Wait for suspect ranks complete: suspect ranks %s", expected)
 
+    def wait_for_aggregation(self, verify_scm=None, verify_nvme=None,
+                             retries=4, interval=30):
+        """Wait for aggregation to finish.
+
+        Args:
+            verify_scm (callable, optional): function(current_scm) to verify scm free space.
+                Defaults to None. Must supply at least one of verify_scm or verify_nvme.
+            verify_nvme (callable, optional): function(current_nvme) to verify nvme free space.
+                Defaults to None. Must supply at least one of verify_scm or verify_nvme.
+            retries (int, optional): number of times to retry. Default is 4.
+            interval (int, optional): seconds to wait before retrying. Default is 60.
+
+        Returns:
+            bool: whether aggregation completed within the time limit
+
+        Raises:
+            ValueError: if neither verify_scm nor verify_nvme are given
+
+        """
+        if verify_scm is None and verify_nvme is None:
+            raise ValueError("verify_scm or verify_nvme is required")
+        for _ in range(retries):
+            current_space = self.get_pool_daos_space()
+            current_free_scm = current_space["s_free"][0]
+            current_free_nvme = current_space["s_free"][1]
+            self.log.info("current_free_scm  = %s", current_free_scm)
+            self.log.info("current_free_nvme = %s", current_free_nvme)
+            if (verify_scm is None or verify_scm(current_free_scm)) and \
+                    (verify_nvme is None or verify_nvme(current_free_scm)):
+                return True
+            sleep(interval)
+
+        return False
+
     def verify_uuid_directory(self, host, scm_mount):
         """Check if pool folder exist on server.