Merge remote-tracking branch 'origin/master' into tanabarr/control-me…

…mfilebytes-mode-mdonssd

Signed-off-by: Tom Nabarro <tom.nabarrointel.com>

.github/workflows/ossf-scorecard.yml

@@ -71,6 +71,6 @@ jobs:
       # Upload the results to GitHub's code scanning dashboard (optional).
       # Commenting out will disable upload of results to your repo's Code Scanning dashboard
       - name: "Upload to code-scanning"
-        uses: github/codeql-action/upload-sarif@48ab28a6f5dbc2a99bf1e0131198dd8f1df78169  # v3.28.0
+        uses: github/codeql-action/upload-sarif@b6a472f63d85b9c78a3ac5e89422239fc15e9b3c  # v3.28.1
         with:
           sarif_file: results.sarif

.github/workflows/trivy.yml

@@ -58,7 +58,7 @@ jobs:
           trivy-config: 'utils/trivy/trivy.yaml'
 
       - name: Upload Trivy scan results to GitHub Security tab
-        uses: github/codeql-action/upload-sarif@48ab28a6f5dbc2a99bf1e0131198dd8f1df78169  # v3.28.0
+        uses: github/codeql-action/upload-sarif@b6a472f63d85b9c78a3ac5e89422239fc15e9b3c  # v3.28.1
         with:
           sarif_file: 'trivy-results.sarif'
 

ci/junit.sh

@@ -24,7 +24,7 @@ report_junit() {
     clush -o '-i ci_key' -l root -w "$nodes" --rcopy "$results"
 
     local results_files
-    results_files=("$results".*)
+    results_files=$(find . -maxdepth 1 -name "$results.*")
 
     if [ ${#results_files[@]} -eq 0 ]; then
         echo "No results found to report as JUnit results"

src/container/srv_target.c

@@ -1,5 +1,7 @@
 /**
  * (C) Copyright 2016-2024 Intel Corporation.
+ * (C) Copyright 2025 Google LLC
+ * (C) Copyright 2025 Hewlett Packard Enterprise Development LP
  *
  * SPDX-License-Identifier: BSD-2-Clause-Patent
  */
@@ -321,7 +323,7 @@ cont_child_aggregate(struct ds_cont_child *cont, cont_aggregate_cb_t agg_cb,
 		     DAOS_FAIL_CHECK(DAOS_FORCE_EC_AGG_PEER_FAIL)))
 		interval = 0;
 	else
-		interval = d_sec2hlc(DAOS_AGG_THRESHOLD);
+		interval = d_sec2hlc(vos_get_agg_gap());
 
 	D_ASSERT(hlc > (interval * 2));
 	/*
@@ -409,6 +411,9 @@ cont_child_aggregate(struct ds_cont_child *cont, cont_aggregate_cb_t agg_cb,
 			DP_CONT(cont->sc_pool->spc_uuid, cont->sc_uuid),
 			tgt_id, epoch_range.epr_lo, epoch_range.epr_hi);
 
+		if (!param->ap_vos_agg)
+			vos_cont_set_mod_bound(cont->sc_hdl, epoch_range.epr_hi);
+
 		flags |= VOS_AGG_FL_FORCE_MERGE;
 		rc = agg_cb(cont, &epoch_range, flags, param);
 		if (rc)
@@ -425,6 +430,9 @@ cont_child_aggregate(struct ds_cont_child *cont, cont_aggregate_cb_t agg_cb,
 		DP_CONT(cont->sc_pool->spc_uuid, cont->sc_uuid),
 		tgt_id, epoch_range.epr_lo, epoch_range.epr_hi);
 
+	if (!param->ap_vos_agg)
+		vos_cont_set_mod_bound(cont->sc_hdl, epoch_range.epr_hi);
+
 	if (dss_xstream_is_busy())
 		flags &= ~VOS_AGG_FL_FORCE_MERGE;
 	rc = agg_cb(cont, &epoch_range, flags, param);
@@ -1607,11 +1615,23 @@ ds_cont_local_open(uuid_t pool_uuid, uuid_t cont_hdl_uuid, uuid_t cont_uuid,
 		 */
 		D_ASSERT(hdl->sch_cont != NULL);
 		D_ASSERT(hdl->sch_cont->sc_pool != NULL);
+
 		hdl->sch_cont->sc_open++;
+		if (hdl->sch_cont->sc_open > 1) {
+			/* If there is an inflight open being stuck, then
+			 * let's retry and wait until it finished.
+			 */
+			if (hdl->sch_cont->sc_open_initializing) {
+				hdl->sch_cont->sc_open--;
+				D_GOTO(err_cont, rc = -DER_AGAIN);
+			}
 
-		if (hdl->sch_cont->sc_open > 1)
-			goto opened;
+			/* Only go through if the 1st open succeeds */
+			if (hdl->sch_cont->sc_props_fetched)
+				goto opened;
+		}
 
+		hdl->sch_cont->sc_open_initializing = 1;
 		if (ds_pool_restricted(hdl->sch_cont->sc_pool->spc_pool, false))
 			goto csum_init;
 
@@ -1646,6 +1666,8 @@ ds_cont_local_open(uuid_t pool_uuid, uuid_t cont_hdl_uuid, uuid_t cont_uuid,
 		rc = ds_cont_csummer_init(hdl->sch_cont);
 		if (rc != 0)
 			D_GOTO(err_dtx, rc);
+
+		hdl->sch_cont->sc_open_initializing = 0;
 	}
 opened:
 	if (cont_hdl != NULL) {
@@ -1663,6 +1685,7 @@ ds_cont_local_open(uuid_t pool_uuid, uuid_t cont_hdl_uuid, uuid_t cont_uuid,
 	dtx_cont_close(hdl->sch_cont, true);
 
 err_cont:
+	hdl->sch_cont->sc_open_initializing = 0;
 	if (daos_handle_is_valid(poh)) {
 		int rc_tmp;
 
@@ -1750,9 +1773,15 @@ ds_cont_tgt_open(uuid_t pool_uuid, uuid_t cont_hdl_uuid,
 	D_DEBUG(DB_TRACE, "open pool/cont/hdl "DF_UUID"/"DF_UUID"/"DF_UUID"\n",
 		DP_UUID(pool_uuid), DP_UUID(cont_uuid), DP_UUID(cont_hdl_uuid));
 
+retry:
 	rc = ds_pool_thread_collective(pool_uuid, PO_COMP_ST_NEW | PO_COMP_ST_DOWN |
 				       PO_COMP_ST_DOWNOUT, cont_open_one, &arg, 0);
-	if (rc != 0)
+	if (rc != 0) {
+		if (rc == -DER_AGAIN) {
+			dss_sleep(50);
+			goto retry;
+		}
+
 		/* Once it exclude the target from the pool, since the target
 		 * might still in the cart group, so IV cont open might still
 		 * come to this target, especially if cont open/close will be
@@ -1762,6 +1791,7 @@ ds_cont_tgt_open(uuid_t pool_uuid, uuid_t cont_hdl_uuid,
 		D_ERROR("open "DF_UUID"/"DF_UUID"/"DF_UUID":"DF_RC"\n",
 			DP_UUID(pool_uuid), DP_UUID(cont_uuid),
 			DP_UUID(cont_hdl_uuid), DP_RC(rc));
+	}
 
 	return rc;
 }

src/dtx/dtx_common.c

@@ -1,5 +1,6 @@
 /**
  * (C) Copyright 2019-2024 Intel Corporation.
+ * (C) Copyright 2025 Hewlett Packard Enterprise Development LP
  *
  * SPDX-License-Identifier: BSD-2-Clause-Patent
  */
@@ -922,6 +923,7 @@ dtx_handle_init(struct dtx_id *dti, daos_handle_t xoh, struct dtx_epoch *epoch,
 	dth->dth_for_migration = (flags & DTX_FOR_MIGRATION) ? 1 : 0;
 	dth->dth_ignore_uncommitted = (flags & DTX_IGNORE_UNCOMMITTED) ? 1 : 0;
 	dth->dth_prepared = (flags & DTX_PREPARED) ? 1 : 0;
+	dth->dth_epoch_owner = (flags & DTX_EPOCH_OWNER) ? 1 : 0;
 	dth->dth_aborted = 0;
 	dth->dth_already = 0;
 	dth->dth_need_validation = 0;

src/dtx/tests/dts_structs.c

@@ -1,5 +1,6 @@
 /**
  * (C) Copyright 2024 Intel Corporation.
+ * (C) Copyright 2025 Hewlett Packard Enterprise Development LP
  *
  * SPDX-License-Identifier: BSD-2-Clause-Patent
  */
@@ -70,8 +71,9 @@ struct_dtx_handle(void **state)
 	SET_BITFIELD_1(dummy, dth_need_validation);
 	SET_BITFIELD_1(dummy, dth_ignore_uncommitted);
 	SET_BITFIELD_1(dummy, dth_local);
+	SET_BITFIELD_1(dummy, dth_epoch_owner);
 	SET_BITFIELD_1(dummy, dth_local_complete);
-	SET_BITFIELD(dummy, padding1, 13);
+	SET_BITFIELD(dummy, padding1, 12);
 
 	SET_FIELD(dummy, dth_dti_cos_count);
 	SET_FIELD(dummy, dth_dti_cos);

src/engine/sched.c

@@ -1,5 +1,6 @@
 /**
  * (C) Copyright 2016-2024 Intel Corporation.
+ * (C) Copyright 2025 Hewlett Packard Enterprise Development LP
  *
  * SPDX-License-Identifier: BSD-2-Clause-Patent
  */
@@ -197,17 +198,6 @@ enum {
 
 static int	sched_policy;
 
-/*
- * Time threshold for giving IO up throttling. If space pressure stays in the
- * highest level for enough long time, we assume that no more space can be
- * reclaimed and choose to give up IO throttling, so that ENOSPACE error could
- * be returned to client earlier.
- *
- * To make time for aggregation reclaiming overwriteen space, this threshold
- * should be longer than the DAOS_AGG_THRESHOLD.
- */
-#define SCHED_DELAY_THRESH	40000	/* msecs */
-
 struct pressure_ratio {
 	unsigned int	pr_free;	/* free space ratio */
 	unsigned int	pr_gc_ratio;	/* CPU percentage for GC & Aggregation */
@@ -781,7 +771,7 @@ check_space_pressure(struct dss_xstream *dx, struct sched_pool_info *spi)
 {
 	struct sched_info	*info = &dx->dx_sched_info;
 	struct vos_pool_space	 vps = { 0 };
-	uint64_t		 scm_left, nvme_left, ne_left, ne_sys;
+	uint64_t                 scm_left, nvme_left, ne_left;
 	struct pressure_ratio	*pr;
 	int			 orig_pressure, rc;
 
@@ -817,12 +807,8 @@ check_space_pressure(struct dss_xstream *dx, struct sched_pool_info *spi)
 	if (vps.vps_ne_total == 0) {
 		ne_left = UINT64_MAX;
 	} else {
-		D_ASSERT(vps.vps_ne_total < SCM_TOTAL(&vps));
-		ne_sys = SCM_SYS(&vps) * vps.vps_ne_total / SCM_TOTAL(&vps);
-		if (vps.vps_ne_free > ne_sys)
-			ne_left = vps.vps_ne_free - ne_sys;
-		else
-			ne_left = 0;
+		ne_left = vps.vps_ne_free;
+		D_ASSERT(ne_left <= vps.vps_ne_total);
 	}
 
 	if (NVME_TOTAL(&vps) == 0)      /* NVMe not enabled */
@@ -943,12 +929,22 @@ is_gc_pending(struct sched_pool_info *spi)
 	return spi->spi_gc_ults && (spi->spi_gc_ults > spi->spi_gc_sleeping);
 }
 
-/* Just run into this space pressure situation recently? */
+/*
+ * Just run into this space pressure situation recently?
+ *
+ * If space pressure stays in the highest level for enough long time, we assume
+ * that no more space can be reclaimed and choose to give up IO throttling, so
+ * that ENOSPACE error could be returned to client earlier.
+ *
+ * To make time for aggregation reclaiming overwriteen space, this threshold
+ * should be longer than VOS aggregation epoch gap against current HLC.
+ */
 static inline bool
 is_pressure_recent(struct sched_info *info, struct sched_pool_info *spi)
 {
 	D_ASSERT(info->si_cur_ts >= spi->spi_pressure_ts);
-	return (info->si_cur_ts - spi->spi_pressure_ts) < SCHED_DELAY_THRESH;
+	return (info->si_cur_ts - spi->spi_pressure_ts) <
+	       (vos_get_agg_gap() + 10) * 1000; /* msecs */
 }
 
 static inline uint64_t

src/include/daos/dtx.h

@@ -1,5 +1,6 @@
 /**
  * (C) Copyright 2019-2023 Intel Corporation.
+ * (C) Copyright 2025 Hewlett Packard Enterprise Development LP
  *
  * SPDX-License-Identifier: BSD-2-Clause-Patent
  */
@@ -27,17 +28,6 @@
 /* The time (in second) threshold for batched DTX commit. */
 #define DTX_COMMIT_THRESHOLD_AGE	10
 
-/*
- * VOS aggregation should try to avoid aggregating in the epoch range where
- * lots of data records are pending to commit, so the aggregation epoch upper
- * bound is: current HLC - (DTX batched commit threshold + buffer period)
- *
- * To avoid conflicting of aggregation vs. transactions, any transactional
- * update/fetch with epoch lower than the aggregation upper bound should be
- * rejected and restarted.
- */
-#define DAOS_AGG_THRESHOLD	(DTX_COMMIT_THRESHOLD_AGE + 10) /* seconds */
-
 enum dtx_target_flags {
 	/* The target only contains read-only operations for the DTX. */
 	DTF_RDONLY			= (1 << 0),

src/include/daos_srv/container.h

@@ -1,5 +1,6 @@
 /*
  * (C) Copyright 2015-2024 Intel Corporation.
+ * (C) Copyright 2025 Google LLC
  *
  * SPDX-License-Identifier: BSD-2-Clause-Patent
  */
@@ -66,20 +67,11 @@ struct ds_cont_child {
 	ABT_cond		 sc_scrub_cond;
 	ABT_cond		 sc_rebuild_cond;
 	ABT_cond		 sc_fini_cond;
-	uint32_t		 sc_dtx_resyncing:1,
-				 sc_dtx_reindex:1,
-				 sc_dtx_reindex_abort:1,
-				 sc_dtx_delay_reset:1,
-				 sc_dtx_registered:1,
-				 sc_props_fetched:1,
-				 sc_stopping:1,
-				 sc_destroying:1,
-				 sc_vos_agg_active:1,
-				 sc_ec_agg_active:1,
-				 /* flag of CONT_CAPA_READ_DATA/_WRITE_DATA disabled */
-				 sc_rw_disabled:1,
-				 sc_scrubbing:1,
-				 sc_rebuilding:1;
+	uint32_t                 sc_dtx_resyncing : 1, sc_dtx_reindex : 1, sc_dtx_reindex_abort : 1,
+	    sc_dtx_delay_reset : 1, sc_dtx_registered : 1, sc_props_fetched : 1, sc_stopping : 1,
+	    sc_destroying : 1, sc_vos_agg_active : 1, sc_ec_agg_active : 1,
+	    /* flag of CONT_CAPA_READ_DATA/_WRITE_DATA disabled */
+	    sc_rw_disabled : 1, sc_scrubbing : 1, sc_rebuilding : 1, sc_open_initializing : 1;
 	uint32_t		 sc_dtx_batched_gen;
 	/* Tracks the schedule request for aggregation ULT */
 	struct sched_request	*sc_agg_req;

src/include/daos_srv/dtx_srv.h

@@ -1,5 +1,6 @@
 /**
  * (C) Copyright 2019-2024 Intel Corporation.
+ * (C) Copyright 2025 Hewlett Packard Enterprise Development LP
  *
  * SPDX-License-Identifier: BSD-2-Clause-Patent
  */
@@ -113,8 +114,10 @@ struct dtx_handle {
 	    dth_ignore_uncommitted                : 1,
 	    /* Local transaction */
 	    dth_local                             : 1,
+	    /* Locally generate the epoch. */
+	    dth_epoch_owner			  : 1,
 	    /* Flag to commit the local transaction */
-	    dth_local_complete : 1, padding1 : 13;
+	    dth_local_complete : 1, padding1 : 12;
 
 	/* The count the DTXs in the dth_dti_cos array. */
 	uint32_t			 dth_dti_cos_count;
@@ -287,6 +290,8 @@ enum dtx_flags {
 	DTX_RELAY = (1 << 10),
 	/** Local transaction */
 	DTX_LOCAL = (1 << 11),
+	/** Locally generate the epoch. */
+	DTX_EPOCH_OWNER = (1 << 12),
 };
 
 void

src/include/daos_srv/vos.h

@@ -1,5 +1,6 @@
 /**
  * (C) Copyright 2015-2024 Intel Corporation.
+ * (C) Copyright 2025 Hewlett Packard Enterprise Development LP
  *
  * SPDX-License-Identifier: BSD-2-Clause-Patent
  */
@@ -939,6 +940,56 @@ vos_update_renew_epoch(daos_handle_t ioh, struct dtx_handle *dth);
 void
 vos_dtx_renew_epoch(struct dtx_handle *dth);
 
+/**
+ * Calculate current locally known stable epoch for the given container.
+ *
+ * \param coh	[IN]	Container open handle
+ *
+ * \return		The epoch on success, negative value if error.
+ */
+daos_epoch_t
+vos_cont_get_local_stable_epoch(daos_handle_t coh);
+
+/**
+ * Get global stable epoch for the given container.
+ *
+ * \param coh	[IN]	Container open handle
+ *
+ * \return		The epoch on success, negative value if error.
+ */
+daos_epoch_t
+vos_cont_get_global_stable_epoch(daos_handle_t coh);
+
+/**
+ * Set global stable epoch for the given container.
+ *
+ * \param coh	[IN]	Container open handle
+ * \param epoch	[IN]	The epoch to be used as the new global stable epoch.
+ *
+ * \return		Zero on success, negative value if error.
+ */
+int
+vos_cont_set_global_stable_epoch(daos_handle_t coh, daos_epoch_t epoch);
+
+/**
+ * Set the lowest allowed modification epoch for the given container.
+ *
+ * \param coh	[IN]	Container open handle
+ * \param epoch	[IN]	The lowest allowed epoch for modification.
+ *
+ * \return		Zero on success, negative value if error.
+ */
+int
+vos_cont_set_mod_bound(daos_handle_t coh, uint64_t epoch);
+
+/**
+ * Query the gap between the max allowed aggregation epoch and current HLC.
+ *
+ * \return		The gap value in seconds.
+ */
+uint32_t
+vos_get_agg_gap(void);
+
 /**
  * Get the recx/epoch list.
  *