DAOS-14909 pool: property for rdb svc_ops KVS entry max age

src/common/prop.c

@@ -422,6 +422,14 @@ daos_prop_valid(daos_prop_t *prop, bool pool, bool input)
 				return false;
 			}
 			break;
+		case DAOS_PROP_PO_SVC_OPS_ENTRY_AGE:
+			val = prop->dpp_entries[i].dpe_val;
+			if ((val < DAOS_PROP_PO_SVC_OPS_ENTRY_AGE_MIN) ||
+			    (val > DAOS_PROP_PO_SVC_OPS_ENTRY_AGE_MAX)) {
+				D_ERROR("invalid svc_ops_entry_age " DF_U64 ".\n", val);
+				return false;
+			}
+			break;
 		/* container-only properties */
 		case DAOS_PROP_CO_LAYOUT_TYPE:
 			val = prop->dpp_entries[i].dpe_val;

src/common/tests_dmg_helpers.c

@@ -681,6 +681,14 @@ dmg_pool_create(const char *dmg_config_file,
 				D_GOTO(out, rc = -DER_NOMEM);
 		}
 
+		entry = daos_prop_entry_get(prop, DAOS_PROP_PO_SVC_OPS_ENTRY_AGE);
+		if (entry != NULL) {
+			args = cmd_push_arg(args, &argcount, "--properties=svc_ops_entry_age:%zu ",
+					    entry->dpe_val);
+			if (args == NULL)
+				D_GOTO(out, rc = -DER_NOMEM);
+		}
+
 		entry = daos_prop_entry_get(prop, DAOS_PROP_PO_SPACE_RB);
 		if (entry != NULL) {
 			args = cmd_push_arg(args, &argcount, "--properties=space_rb:%zu ",

src/control/lib/daos/pool_cont_prop.go

@@ -126,8 +126,9 @@ const (
 	//PoolPropertyPerfDomain is pool performance domain
 	PoolPropertyPerfDomain = C.DAOS_PROP_PO_PERF_DOMAIN
 	//PoolPropertyReintMode is pool reintegration mode
-	PoolPropertyReintMode     = C.DAOS_PROP_PO_REINT_MODE
-	PoolPropertySvcOpsEnabled = C.DAOS_PROP_PO_SVC_OPS_ENABLED
+	PoolPropertyReintMode      = C.DAOS_PROP_PO_REINT_MODE
+	PoolPropertySvcOpsEnabled  = C.DAOS_PROP_PO_SVC_OPS_ENABLED
+	PoolPropertySvcOpsEntryAge = C.DAOS_PROP_PO_SVC_OPS_ENTRY_AGE
 )
 
 const (
@@ -186,6 +187,8 @@ const (
 	PoolSvcRedunFacMax = C.DAOS_PROP_PO_SVC_REDUN_FAC_MAX
 	// PoolSvcRedunFacDefault defines the default value of PoolPropertySvcRedunFac.
 	PoolSvcRedunFacDefault = C.DAOS_PROP_PO_SVC_REDUN_FAC_DEFAULT
+	PoolSvcOpsEntryAgeMin  = C.DAOS_PROP_PO_SVC_OPS_ENTRY_AGE_MIN
+	PoolSvcOpsEntryAgeMax  = C.DAOS_PROP_PO_SVC_OPS_ENTRY_AGE_MAX
 )
 
 const (

src/control/lib/daos/pool_property.go

@@ -120,6 +120,31 @@ func PoolProperties() PoolPropertyMap {
 				valueMarshaler: numericMarshaler,
 			},
 		},
+		"svc_ops_entry_age": {
+			Property: PoolProperty{
+				Number:      PoolPropertySvcOpsEntryAge,
+				Description: "Metadata duplicate operations KVS max entry age, in seconds",
+				valueHandler: func(s string) (*PoolPropertyValue, error) {
+					oeErr := errors.Errorf("invalid svc_ops_entry_age %s (valid values: %d-%d)", s, PoolSvcOpsEntryAgeMin, PoolSvcOpsEntryAgeMax)
+					oeVal, err := strconv.ParseUint(s, 10, 32)
+					if err != nil {
+						return nil, oeErr
+					}
+					if oeVal < PoolSvcOpsEntryAgeMin || oeVal > PoolSvcOpsEntryAgeMax {
+						return nil, errors.Wrap(oeErr, "value supplied is out of range")
+					}
+					return &PoolPropertyValue{oeVal}, nil
+				},
+				valueStringer: func(v *PoolPropertyValue) string {
+					n, err := v.GetNumber()
+					if err != nil {
+						return "not set"
+					}
+					return fmt.Sprintf("%d", n)
+				},
+				valueMarshaler: numericMarshaler,
+			},
+		},
 		"label": {
 			Property: PoolProperty{
 				Number:      PoolPropertyLabel,

src/include/daos/pool.h

@@ -51,7 +51,8 @@
 #define DAOS_PO_QUERY_PROP_CHECKPOINT_THRESH    (1ULL << (PROP_BIT_START + 23))
 #define DAOS_PO_QUERY_PROP_REINT_MODE		(1ULL << (PROP_BIT_START + 24))
 #define DAOS_PO_QUERY_PROP_SVC_OPS_ENABLED      (1ULL << (PROP_BIT_START + 25))
-#define DAOS_PO_QUERY_PROP_BIT_END              41
+#define DAOS_PO_QUERY_PROP_SVC_OPS_ENTRY_AGE    (1ULL << (PROP_BIT_START + 26))
+#define DAOS_PO_QUERY_PROP_BIT_END              42
 
 #define DAOS_PO_QUERY_PROP_ALL                                                                     \
 	(DAOS_PO_QUERY_PROP_LABEL | DAOS_PO_QUERY_PROP_SPACE_RB | DAOS_PO_QUERY_PROP_SELF_HEAL |   \
@@ -65,7 +66,7 @@
 	 DAOS_PO_QUERY_PROP_OBJ_VERSION | DAOS_PO_QUERY_PROP_PERF_DOMAIN |                         \
 	 DAOS_PO_QUERY_PROP_CHECKPOINT_MODE | DAOS_PO_QUERY_PROP_CHECKPOINT_FREQ |                 \
 	 DAOS_PO_QUERY_PROP_CHECKPOINT_THRESH | DAOS_PO_QUERY_PROP_REINT_MODE |                    \
-	 DAOS_PO_QUERY_PROP_SVC_OPS_ENABLED)
+	 DAOS_PO_QUERY_PROP_SVC_OPS_ENABLED | DAOS_PO_QUERY_PROP_SVC_OPS_ENTRY_AGE)
 
 /*
  * Version 1 corresponds to 2.2 (aggregation optimizations)

src/include/daos_prop.h

@@ -137,6 +137,8 @@ enum daos_pool_props {
 	DAOS_PROP_PO_REINT_MODE,
 	/** Metadata duplicate operations detection enabled (1) or disabled (0) */
 	DAOS_PROP_PO_SVC_OPS_ENABLED,
+	/** Metadata duplicate operations SVC_OPS KVS max entry age (seconds), default 300 */
+	DAOS_PROP_PO_SVC_OPS_ENTRY_AGE,
 	DAOS_PROP_PO_MAX,
 };
 
@@ -246,13 +248,16 @@ enum {
 };
 
 #define DAOS_PROP_PO_CHECKPOINT_MODE_DEFAULT   DAOS_CHECKPOINT_TIMED
-#define DAOS_PROP_PO_CHECKPOINT_FREQ_DEFAULT   5  /* 5 seconds */
-#define DAOS_PROP_PO_CHECKPOINT_FREQ_MIN       1  /* 1 seconds */
+#define DAOS_PROP_PO_CHECKPOINT_FREQ_DEFAULT   5         /* 5 seconds */
+#define DAOS_PROP_PO_CHECKPOINT_FREQ_MIN       1         /* 1 seconds */
 #define DAOS_PROP_PO_CHECKPOINT_FREQ_MAX       (1 << 20) /* 1 million seconds */
-#define DAOS_PROP_PO_CHECKPOINT_THRESH_DEFAULT 50 /* 50 % WAL capacity */
-#define DAOS_PROP_PO_CHECKPOINT_THRESH_MAX     75 /* 75 % WAL capacity */
-#define DAOS_PROP_PO_CHECKPOINT_THRESH_MIN     10 /* 10 % WAL capacity */
-#define DAOS_PROP_PO_SVC_OPS_ENABLED_DEFAULT   1  /* true: enabled by default */
+#define DAOS_PROP_PO_CHECKPOINT_THRESH_DEFAULT 50        /* 50 % WAL capacity */
+#define DAOS_PROP_PO_CHECKPOINT_THRESH_MAX     75        /* 75 % WAL capacity */
+#define DAOS_PROP_PO_CHECKPOINT_THRESH_MIN     10        /* 10 % WAL capacity */
+#define DAOS_PROP_PO_SVC_OPS_ENABLED_DEFAULT   1         /* true: enabled by default */
+#define DAOS_PROP_PO_SVC_OPS_ENTRY_AGE_DEFAULT 300       /* 300 seconds */
+#define DAOS_PROP_PO_SVC_OPS_ENTRY_AGE_MIN     150       /* 150 seconds */
+#define DAOS_PROP_PO_SVC_OPS_ENTRY_AGE_MAX     600       /* 600 seconds */
 
 /** self healing strategy bits */
 #define DAOS_SELF_HEAL_AUTO_EXCLUDE	(1U << 0)

src/pool/rpc.c

@@ -357,6 +357,9 @@ pool_query_bits(daos_pool_info_t *po_info, daos_prop_t *prop)
 		case DAOS_PROP_PO_SVC_OPS_ENABLED:
 			bits |= DAOS_PO_QUERY_PROP_SVC_OPS_ENABLED;
 			break;
+		case DAOS_PROP_PO_SVC_OPS_ENTRY_AGE:
+			bits |= DAOS_PO_QUERY_PROP_SVC_OPS_ENTRY_AGE;
+			break;
 		default:
 			D_ERROR("ignore bad dpt_type %d.\n", entry->dpe_type);
 			break;

src/pool/srv_internal.h

@@ -93,6 +93,7 @@ struct pool_iv_prop {
 	uint32_t	pip_perf_domain;
 	uint32_t	pip_reint_mode;
 	uint32_t         pip_svc_ops_enabled;
+	uint32_t         pip_svc_ops_entry_age;
 	char		pip_iv_buf[0];
 };
 

src/pool/srv_iv.c

@@ -223,6 +223,9 @@ pool_iv_prop_l2g(daos_prop_t *prop, struct pool_iv_prop *iv_prop)
 		case DAOS_PROP_PO_SVC_OPS_ENABLED:
 			iv_prop->pip_svc_ops_enabled = prop_entry->dpe_val;
 			break;
+		case DAOS_PROP_PO_SVC_OPS_ENTRY_AGE:
+			iv_prop->pip_svc_ops_entry_age = prop_entry->dpe_val;
+			break;
 		default:
 			D_ASSERTF(0, "bad dpe_type %d\n", prop_entry->dpe_type);
 			break;
@@ -366,6 +369,9 @@ pool_iv_prop_g2l(struct pool_iv_prop *iv_prop, daos_prop_t *prop)
 		case DAOS_PROP_PO_SVC_OPS_ENABLED:
 			prop_entry->dpe_val = iv_prop->pip_svc_ops_enabled;
 			break;
+		case DAOS_PROP_PO_SVC_OPS_ENTRY_AGE:
+			prop_entry->dpe_val = iv_prop->pip_svc_ops_entry_age;
+			break;
 		default:
 			D_ASSERTF(0, "bad dpe_type %d\n", prop_entry->dpe_type);
 			break;

src/pool/srv_layout.c

@@ -158,6 +158,10 @@ struct daos_prop_entry pool_prop_entries_default[DAOS_PROP_PO_NUM] = {
     {
 	.dpe_type = DAOS_PROP_PO_SVC_OPS_ENABLED,
 	.dpe_val  = DAOS_PROP_PO_SVC_OPS_ENABLED_DEFAULT,
+    },
+    {
+	.dpe_type = DAOS_PROP_PO_SVC_OPS_ENTRY_AGE,
+	.dpe_val  = DAOS_PROP_PO_SVC_OPS_ENTRY_AGE_DEFAULT,
     }};
 
 daos_prop_t pool_prop_default = {

src/pool/srv_pool.c

@@ -42,10 +42,7 @@
 #define DAOS_POOL_GLOBAL_VERSION_WITH_SVC_OPS_KVS 3
 
 #define PS_OPS_PER_SEC                            4096
-#define DEFAULT_PS_OPS_ENTRY_AGE_SEC              300ULL
-#define MIN_PS_OPS_ENTRY_AGE_SEC                  150
-#define MAX_PS_OPS_ENTRY_AGE_SEC                  600
-#define DAOS_MD_OPS_AGE_SEC_ENV                   "DAOS_MD_OPS_AGE_SEC"
+
 /*
  * Return the corresponding VOS pool DF version or 0 if pool_global_version is
  * not supported.
@@ -374,6 +371,7 @@ pool_prop_default_copy(daos_prop_t *prop_def, daos_prop_t *prop)
 		case DAOS_PROP_PO_SVC_REDUN_FAC:
 		case DAOS_PROP_PO_PERF_DOMAIN:
 		case DAOS_PROP_PO_SVC_OPS_ENABLED:
+		case DAOS_PROP_PO_SVC_OPS_ENTRY_AGE:
 			entry_def->dpe_val = entry->dpe_val;
 			break;
 		case DAOS_PROP_PO_POLICY:
@@ -682,6 +680,13 @@ pool_prop_write(struct rdb_tx *tx, const rdb_path_t *kvs, daos_prop_t *prop)
 			if (rc)
 				return rc;
 			break;
+		case DAOS_PROP_PO_SVC_OPS_ENTRY_AGE:
+			val32 = entry->dpe_val;
+			d_iov_set(&value, &val32, sizeof(val32));
+			rc = rdb_tx_update(tx, kvs, &ds_pool_prop_svc_ops_age, &value);
+			if (rc)
+				return rc;
+			break;
 		default:
 			D_ERROR("bad dpe_type %d.\n", entry->dpe_type);
 			return -DER_INVAL;
@@ -695,25 +700,6 @@ pool_prop_write(struct rdb_tx *tx, const rdb_path_t *kvs, daos_prop_t *prop)
 	return rc;
 }
 
-static uint32_t
-get_svc_ops_age(void)
-{
-	const uint32_t ops_age_default = DEFAULT_PS_OPS_ENTRY_AGE_SEC;
-	char          *v;
-	int            n;
-
-	v = getenv(DAOS_MD_OPS_AGE_SEC_ENV); /* in MB */
-	if (v == NULL)
-		return ops_age_default;
-	n = atoi(v);
-	if ((n < MIN_PS_OPS_ENTRY_AGE_SEC) || (n > MAX_PS_OPS_ENTRY_AGE_SEC)) {
-		D_ERROR("metadata ps ops age %d out of range %u..%u; using %u sec\n", n,
-			MIN_PS_OPS_ENTRY_AGE_SEC, MAX_PS_OPS_ENTRY_AGE_SEC, ops_age_default);
-		return ops_age_default;
-	}
-	return n;
-}
-
 static int
 init_pool_metadata(struct rdb_tx *tx, const rdb_path_t *kvs, uint32_t nnodes, const char *group,
 		   const d_rank_list_t *ranks, daos_prop_t *prop, uint32_t ndomains,
@@ -729,7 +715,7 @@ init_pool_metadata(struct rdb_tx *tx, const rdb_path_t *kvs, uint32_t nnodes, co
 	uint32_t		upgrade_global_version = DAOS_POOL_GLOBAL_VERSION;
 	uint32_t                svc_ops_enabled        = 1;
 	/* max number of entries in svc_ops KVS: equivalent of max age (sec) x PS_OPS_PER_SEC */
-	uint32_t                svc_ops_age;
+	uint32_t                svc_ops_age = DAOS_PROP_PO_SVC_OPS_ENTRY_AGE_DEFAULT;
 	uint32_t                svc_ops_max;
 	uint32_t                svc_ops_num;
 	uint64_t                rdb_size;
@@ -841,10 +827,12 @@ init_pool_metadata(struct rdb_tx *tx, const rdb_path_t *kvs, uint32_t nnodes, co
 		goto out_map_buf;
 	}
 
-	/* Maximum number of RPCs that may be kept in svc_ops.
-	 * Default: number of RPCs equivalent to PS_OPS_PER_SEC x DEFAULT_PS_OPS_ENTRY_AGE_SEC.
+	/* Maximum number of RPCs that may be kept in svc_ops, from SVC_OPS_ENTRY_AGE property.
+	 * Default: PS_OPS_PER_SEC x DEFAULT_SVC_OPS_ENTRY_AGE_SEC.
 	 */
-	svc_ops_age = get_svc_ops_age();
+	entry = daos_prop_entry_get(prop, DAOS_PROP_PO_SVC_OPS_ENTRY_AGE);
+	if (entry)
+		svc_ops_age = entry->dpe_val;
 	svc_ops_max = PS_OPS_PER_SEC * svc_ops_age;
 	svc_ops_num = 0;
 	d_iov_set(&value, &svc_ops_age, sizeof(svc_ops_age));
@@ -2775,6 +2763,27 @@ pool_prop_read(struct rdb_tx *tx, const struct pool_svc *svc, uint64_t bits,
 		idx++;
 	}
 
+	if (bits & DAOS_PO_QUERY_PROP_SVC_OPS_ENTRY_AGE) {
+		d_iov_set(&value, &val32, sizeof(val32));
+		rc = rdb_tx_lookup(tx, &svc->ps_root, &ds_pool_prop_svc_ops_age, &value);
+		if (rc == -DER_NONEXIST && global_ver < DAOS_POOL_GLOBAL_VERSION_WITH_SVC_OPS_KVS) {
+			/* needs to be upgraded */
+			rc    = 0;
+			val32 = 0;
+			prop->dpp_entries[idx].dpe_flags |= DAOS_PROP_ENTRY_NOT_SET;
+		} else if (rc != 0) {
+			DL_ERROR(rc, DF_UUID ": DAOS_PROP_PO_SVC_OPS_ENTRY_AGE missing from pool",
+				 DP_UUID(svc->ps_uuid));
+			D_GOTO(out_prop, rc);
+		}
+		if (rc != 0)
+			D_GOTO(out_prop, rc);
+		D_ASSERT(idx < nr);
+		prop->dpp_entries[idx].dpe_type = DAOS_PROP_PO_SVC_OPS_ENTRY_AGE;
+		prop->dpp_entries[idx].dpe_val  = val32;
+		idx++;
+	}
+
 	*prop_out = prop;
 	return 0;
 
@@ -4518,6 +4527,7 @@ ds_pool_query_handler(crt_rpc_t *rpc, int handler_version)
 			case DAOS_PROP_PO_CHECKPOINT_THRESH:
 			case DAOS_PROP_PO_REINT_MODE:
 			case DAOS_PROP_PO_SVC_OPS_ENABLED:
+			case DAOS_PROP_PO_SVC_OPS_ENTRY_AGE:
 				if (entry->dpe_val != iv_entry->dpe_val) {
 					D_ERROR("type %d mismatch "DF_U64" - "
 						DF_U64".\n", entry->dpe_type,
@@ -5981,6 +5991,11 @@ ds_pool_svc_set_prop(uuid_t pool_uuid, d_rank_list_t *ranks, daos_prop_t *prop)
 		D_GOTO(out, rc = -DER_NO_PERM);
 	}
 
+	if (daos_prop_entry_get(prop, DAOS_PROP_PO_SVC_OPS_ENTRY_AGE)) {
+		D_ERROR("Can't set pool svc_ops_entry_age on existing pool.\n");
+		D_GOTO(out, rc = -DER_NO_PERM);
+	}
+
 	/* Disallow to begin with; will support in the future. */
 	if (daos_prop_entry_get(prop, DAOS_PROP_PO_SVC_REDUN_FAC)) {
 		D_ERROR(DF_UUID ": cannot set pool service redundancy factor on existing pool\n",

src/tests/suite/daos_pool.c

@@ -499,6 +499,7 @@ pool_properties(void **state)
 	daos_prop_t		*prop_query;
 	struct daos_prop_entry	*entry;
 	daos_pool_info_t	 info = {0};
+	const uint64_t           svc_ops_age = 180;
 	int			 rc;
 	char			*expected_owner;
 	char			*expected_group;
@@ -510,7 +511,7 @@ pool_properties(void **state)
 			SMALL_POOL_SIZE, 0, NULL);
 	assert_rc_equal(rc, 0);
 
-	prop = daos_prop_alloc(4);
+	prop = daos_prop_alloc(5);
 	/* label - set arg->pool_label to use daos_pool_connect() */
 	prop->dpp_entries[0].dpe_type = DAOS_PROP_PO_LABEL;
 	D_STRNDUP_S(prop->dpp_entries[0].dpe_str, label);
@@ -524,8 +525,11 @@ pool_properties(void **state)
 	prop->dpp_entries[2].dpe_type = DAOS_PROP_PO_SVC_OPS_ENABLED;
 	prop->dpp_entries[2].dpe_val  = 0; /* disabled */
 
-	prop->dpp_entries[3].dpe_type = DAOS_PROP_PO_SPACE_RB;
-	prop->dpp_entries[3].dpe_val  = space_rb;
+	prop->dpp_entries[3].dpe_type = DAOS_PROP_PO_SVC_OPS_ENTRY_AGE;
+	prop->dpp_entries[3].dpe_val  = svc_ops_age; /* seconds */
+
+	prop->dpp_entries[4].dpe_type = DAOS_PROP_PO_SPACE_RB;
+	prop->dpp_entries[4].dpe_val  = space_rb;
 
 	while (!rc && arg->setup_state != SETUP_POOL_CONNECT)
 		rc = test_setup_next_step((void **)&arg, NULL, prop, NULL);
@@ -559,6 +563,11 @@ pool_properties(void **state)
 		fail_msg("svc_ops_enabled verification failed.\n");
 	}
 
+	entry = daos_prop_entry_get(prop_query, DAOS_PROP_PO_SVC_OPS_ENTRY_AGE);
+	if (entry == NULL || (entry->dpe_val != svc_ops_age)) {
+		fail_msg("svc_ops_entry_age verification failed.\n");
+	}
+
 	entry = daos_prop_entry_get(prop_query, DAOS_PROP_PO_SPACE_RB);
 	if (entry == NULL || entry->dpe_val != space_rb) {
 		fail_msg("space_rb verification failed.\n");