DAOS-16809 vos: container based stable epoch

For the purpose of efficient calculating container based local stable epoch,
we will maintain some kind of sorted list for active DTX entries with epoch
order. But consider related overhead, it is not easy to maintain a strictly
sorted list for all active DTX entries. For the DTX which leader resides on
current target, its epoch is already sorted when generate on current engine.
So the main difficulty is for those DTX entries which leaders are on remote
targets.

On the other hand, the local stable epoch is mainly used to generate global
stable epoch that is for incremental reintegration. In fact, we do not need
a very accurate global stable epoch for incremental reintegration. It means
that it is no matter (or non-fatal) if the calculated stable epoch is a bit
smaller than the real case. For example, seconds error for the stable epoch
almost can be ignored if we compare such overhead with rebuilding the whole
target from scratch. So for the DTX entry which leader is on remomte target,
we will maintain it in the list with relative incremental trend based on the
epoch instead of strict sorting the epoch. We introduce an O(1) algorithm to
handle such unsroted DTX entries list for calculating local stable epoch.

Main VOS APIs for the stable epoch:

/* Calculate current locally known stable epoch for the given container. */
daos_epoch_t vos_cont_get_local_stable_epoch(daos_handle_t coh);

/* Get global stable epoch for the given container. */
daos_epoch_t vos_cont_get_global_stable_epoch(daos_handle_t coh);

/* Set global stable epoch for the given container. */
int vos_cont_set_global_stable_epoch(daos_handle_t coh, daos_epoch_t epoch);

Another important enhancement in the patch is about handling potential
conflict between EC/VOS aggregation and delayed modification with very
old epoch.

For standalone transaction, when it is started on the DTX leader, its epoch
is generated by the leader, then the modification RPC will be forwarded to
other related non-leader(s). If the forwarded RPC is delayed for some reason,
such as network congestion or system busy on the non-leader, as to the epoch
for such transaction becomes very old (exceed related threshold), as to VOS
aggregation may has already aggregated related epoch rang. Under such case,
the non-leader will reject such modification to avoid data lost/corruption.

For distributed transaction, if there is no read (fetch, query, enumerate,
and so on) before client tx_commit, then related DTX leader will generate
epoch for the transaction after client commit_tx. Then it will be the same
as above standalone transaction for epoch handling.

If the distributed transaction involves some read before client commit_tx,
its epoch will be generated by the first accessed engine for read. If the
transaction takes too long time after that, then when client commit_tx, its
epoch may become very old as to related DTX leader will have to reject the
transaction to avoid above mentioned conflict. And even if the DTX leader
did not reject the transaction, some non-leader may also reject it because
of the very old epoch. So it means that under such framework, the life for
a distributed transaction cannot be too long. That can be adjusted via the
server side environment variable DAOS_VOS_AGG_GAP. The defaule value is 60
seconds.

NOTE: EC/VOS aggregation should avoid aggregating in the epoch range where
      lots of data records are pending to commit, so the aggregation epoch
      upper bound is 'current HLC - vos_agg_gap'.

Signed-off-by: Fan Yong <[email protected]>

src/container/srv_target.c

@@ -321,7 +321,7 @@ 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 = cont->sc_agg_eph_gap;
 
 	D_ASSERT(hlc > (interval * 2));
 	/*
@@ -409,6 +409,9 @@ 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 +428,9 @@ 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);

src/dtx/dtx_common.c

@@ -922,6 +922,7 @@ 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;
@@ -1853,6 +1854,8 @@ dtx_cont_register(struct ds_cont_child *cont)
 		D_GOTO(out, rc = -DER_NOMEM);
 	}
 
+	cont->sc_agg_eph_gap = d_sec2hlc(vos_get_agg_gap());
+
 	ds_cont_child_get(cont);
 	dbca->dbca_refs = 0;
 	dbca->dbca_cont = cont;

src/dtx/tests/dts_structs.c

@@ -70,8 +70,9 @@ 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

@@ -197,17 +197,6 @@ 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 */
@@ -943,12 +932,21 @@ 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 with 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) < info->si_agg_gap;
 }
 
 static inline uint64_t
@@ -2256,6 +2254,8 @@ sched_run(ABT_sched sched)
 		return;
 	}
 
+	dx->dx_sched_info.si_agg_gap = (vos_get_agg_gap() + 10) * 1000; /* msecs */
+
 	while (1) {
 		/* Try to pick network poll ULT */
 		pool = pools[DSS_POOL_NET_POLL];

src/engine/srv_internal.h

@@ -61,6 +61,7 @@ struct sched_info {
 	/* Number of kicked requests for each type in current cycle */
 	uint32_t		 si_kicked_req_cnt[SCHED_REQ_MAX];
 	unsigned int		 si_stop:1;
+	uint64_t		 si_agg_gap;
 };
 
 struct mem_stats {

src/include/daos/dtx.h

@@ -1,5 +1,5 @@
 /**
- * (C) Copyright 2019-2023 Intel Corporation.
+ * (C) Copyright 2019-2024 Intel Corporation.
  *
  * SPDX-License-Identifier: BSD-2-Clause-Patent
  */
@@ -27,17 +27,6 @@
 /* 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

@@ -129,6 +129,12 @@ struct ds_cont_child {
 	 */
 	uint64_t		sc_ec_update_timestamp;
 
+	/*
+	 * The gap between the max allowed aggregation epoch and current HLC. The modification
+	 * with older epoch out of range may cause conflict with aggregation as to be rejected.
+	 */
+	uint64_t		 sc_agg_eph_gap;
+
 	/* The objects with committable DTXs in DRAM. */
 	daos_handle_t		 sc_dtx_cos_hdl;
 	/* The DTX COS-btree. */

src/include/daos_srv/dtx_srv.h

@@ -113,8 +113,10 @@ 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 +289,8 @@ 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

@@ -938,6 +938,56 @@ 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.
  *

src/include/daos_srv/vos_types.h

@@ -58,6 +58,8 @@ enum dtx_entry_flags {
 	 * on all yet, need to be re-committed.
 	 */
 	DTE_PARTIAL_COMMITTED	= (1 << 5),
+	/* The DTX epoch is sorted locally. */
+	DTE_EPOCH_SORTED	= (1 << 6),
 };
 
 struct dtx_entry {

src/object/srv_obj.c

@@ -2896,8 +2896,10 @@ ds_obj_rw_handler(crt_rpc_t *rpc)
 
 	rc = process_epoch(&orw->orw_epoch, &orw->orw_epoch_first,
 			   &orw->orw_flags);
-	if (rc == PE_OK_LOCAL)
+	if (rc == PE_OK_LOCAL) {
 		orw->orw_flags &= ~ORF_EPOCH_UNCERTAIN;
+		dtx_flags |= DTX_EPOCH_OWNER;
+	}
 
 	if (obj_rpc_is_fetch(rpc)) {
 		struct dtx_handle	*dth;
@@ -3856,8 +3858,10 @@ ds_obj_punch_handler(crt_rpc_t *rpc)
 
 	rc = process_epoch(&opi->opi_epoch, NULL /* epoch_first */,
 			   &opi->opi_flags);
-	if (rc == PE_OK_LOCAL)
+	if (rc == PE_OK_LOCAL) {
 		opi->opi_flags &= ~ORF_EPOCH_UNCERTAIN;
+		dtx_flags |= DTX_EPOCH_OWNER;
+	}
 
 	version = opi->opi_map_ver;
 	max_ver = opi->opi_map_ver;
@@ -5110,6 +5114,7 @@ ds_obj_dtx_leader(struct daos_cpd_args *dca)
 			   &dcsh->dcsh_epoch.oe_first,
 			   &dcsh->dcsh_epoch.oe_rpc_flags);
 	if (rc == PE_OK_LOCAL) {
+		dtx_flags |= DTX_EPOCH_OWNER;
 		/*
 		 * In this case, writes to local RDGs can use the chosen epoch
 		 * without any uncertainty. This optimization is left to future
@@ -5701,8 +5706,10 @@ ds_obj_coll_punch_handler(crt_rpc_t *rpc)
 
 	if (ocpi->ocpi_flags & ORF_LEADER) {
 		rc = process_epoch(&ocpi->ocpi_epoch, NULL /* epoch_first */, &ocpi->ocpi_flags);
-		if (rc == PE_OK_LOCAL)
+		if (rc == PE_OK_LOCAL) {
 			ocpi->ocpi_flags &= ~ORF_EPOCH_UNCERTAIN;
+			dtx_flags |= DTX_EPOCH_OWNER;
+		}
 	} else if (dct_nr == 1) {
 		rc = obj_coll_local(rpc, dcts[0].dct_shards, dce, &version, &ioc, NULL,
 				    odm->odm_mbs, obj_coll_tgt_punch);

src/tests/ftest/util/server_utils_params.py

@@ -436,6 +436,7 @@ class EngineYamlParameters(YamlParameters):
             "D_LOG_FILE_APPEND_PID=1",
             "DAOS_POOL_RF=4",
             "CRT_EVENT_DELAY=1",
+            "DAOS_VOS_AGG_GAP=20",
             "COVFILE=/tmp/test.cov"],
         "ofi+tcp": [],
         "ofi+tcp;ofi_rxm": [],

src/vos/tests/vts_dtx.c

@@ -66,6 +66,7 @@ vts_dtx_begin(const daos_unit_oid_t *oid, daos_handle_t coh, daos_epoch_t epoch,
 	dth->dth_for_migration = 0;
 	dth->dth_ignore_uncommitted = 0;
 	dth->dth_prepared = 0;
+	dth->dth_epoch_owner = 0;
 	dth->dth_aborted = 0;
 	dth->dth_already = 0;
 	dth->dth_need_validation = 0;

src/vos/vos_common.c

@@ -620,6 +620,42 @@ struct dss_module_key vos_module_key = {
 
 daos_epoch_t	vos_start_epoch = DAOS_EPOCH_MAX;
 
+/*
+ * For standalone transaction, when it is started on the DTX leader, its epoch
+ * is generated by the leader, then the modification RPC will be forwarded to
+ * other related non-leader(s). If the forwarded RPC is delayed for some reason,
+ * such as network congestion or system busy on the non-leader, as to the epoch
+ * for such transaction becomes very old (exceed related threshold), as to VOS
+ * aggregation may has already aggregated related epoch rang. Under such case,
+ * the non-leader will reject such modification to avoid data lost/corruption.
+ *
+ * For distributed transaction, if there is no read (fetch, query, enumerate,
+ * and so on) before client tx_commit, then related DTX leader will generate
+ * epoch for the transaction after client commit_tx. Then it will be the same
+ * as above standalone transaction for epoch handling.
+ *
+ * If the distributed transaction involves some read before client commit_tx,
+ * its epoch will be generated by the first accessed engine for read. If the
+ * transaction takes too long time after that, then when client commit_tx, its
+ * epoch may become very old as to related DTX leader will have to reject the
+ * transaction to avoid above mentioned conflict. And even if the DTX leader
+ * did not reject the transaction, some non-leader may also reject it because
+ * of the very old epoch. So it means that under such framework, the life for
+ * a distributed transaction cannot be too long. That can be adjusted via the
+ * server side environment variable DAOS_VOS_AGG_GAP.
+ *
+ * NOTE: EC/VOS aggregation should avoid aggregating in the epoch range where
+ *	 lots of data records are pending to commit, so the aggregation epoch
+ *	 upper bound is 'current HLC - vos_agg_gap'.
+ */
+uint32_t	vos_agg_gap;
+
+uint32_t
+vos_get_agg_gap(void)
+{
+	return vos_agg_gap;
+}
+
 static int
 vos_mod_init(void)
 {
@@ -679,6 +715,15 @@ vos_mod_init(void)
 	d_getenv_bool("DAOS_DKEY_PUNCH_PROPAGATE", &vos_dkey_punch_propagate);
 	D_INFO("DKEY punch propagation is %s\n", vos_dkey_punch_propagate ? "enabled" : "disabled");
 
+	vos_agg_gap = VOS_AGG_GAP_DEF;
+	d_getenv_uint("DAOS_VOS_AGG_GAP", &vos_agg_gap);
+	if (vos_agg_gap < VOS_AGG_GAP_MIN || vos_agg_gap > VOS_AGG_GAP_MAX) {
+		D_WARN("Invalid DAOS_VOS_AGG_GAP value, "
+		       "valid range [%u, %u], set it as default %u (second)\n",
+		       VOS_AGG_GAP_MIN, VOS_AGG_GAP_MAX, VOS_AGG_GAP_DEF);
+		vos_agg_gap = VOS_AGG_GAP_DEF;
+	}
+	D_INFO("Set DAOS VOS aggregation gap as %u (second)\n", vos_agg_gap);
 
 	return rc;
 }