Merge remote-tracking branch 'origin/release/2.4' into dbohning/daos-…

…13555-2.4

Test-tag: test_always_passes,vm
Skip-unit-tests: true
Skip-fault-injection-test: true

ci/docker_nlt.sh

@@ -22,7 +22,7 @@ pushd "$TMP_DIR"
 set +e
 
 sudo --preserve-env=VIRTUAL_ENV,PATH ./node_local_test.py \
- --no-root --memcheck no --system-ram-reserved 32 --server-debug WARN "$@"
+ --no-root --memcheck no --system-ram-reserved 48 --server-debug WARN "$@"
 
 RC=$?
 set -e

docs/admin/env_variables.md

@@ -68,6 +68,7 @@ Environment variables in this section only apply to the client side.
 |Variable |Description|
 |-------------------------|-----------|
 |FI\_MR\_CACHE\_MAX\_COUNT|Enable MR (Memory Registration) caching in OFI layer. Recommended to be set to 0 (disable) when CRT\_DISABLE\_MEM\_PIN is NOT set to 1. INTEGER. Default to unset.|
+|D\_POLL\_TIMEOUT|Polling timeout passed to network progress for synchronous operations. Default to 0 (busy polling), value in micro-seconds otherwise.|
 
 
 ## Debug System (Client & Server)

src/cart/crt_hg.c

@@ -37,10 +37,14 @@ struct crt_na_dict crt_na_dict[] = {
  .nad_str = "ofi+gni",
  .nad_contig_eps = true,
  .nad_port_bind = false,
+ }, {
+ .nad_type = CRT_PROV_OFI_TCP,
+ .nad_str = "ofi+tcp",
+ .nad_contig_eps = true,
+ .nad_port_bind = true,
  }, {
  .nad_type = CRT_PROV_OFI_TCP_RXM,
  .nad_str = "ofi+tcp;ofi_rxm",
- .nad_alt_str = "ofi+tcp",
  .nad_contig_eps = true,
  .nad_port_bind = true,
  }, {
@@ -655,7 +659,7 @@ crt_get_opx_info_string(char *provider, char *domain, char *ip,
  D_GOTO(out, rc = -DER_INVAL);
  }
 
- strcpy(domain_name, domain);
+ strncpy(domain_name, domain, sizeof(domain_name) - 1);
  strtok_r(domain_name, &domain[delimiter], &hfi_str);
  hfi = (unsigned int)strtoul(hfi_str, NULL, 10);
 

src/cart/crt_hg.h

@@ -31,6 +31,35 @@ struct crt_rpc_priv;
 struct crt_common_hdr;
 struct crt_corpc_hdr;
 
+/**
+ * Enumeration specifying providers supported by the library
+ */
+typedef enum {
+ CRT_PROV_SM = 0,
+ CRT_PROV_OFI_SOCKETS,
+ CRT_PROV_OFI_VERBS_RXM,
+ CRT_PROV_OFI_GNI,
+ CRT_PROV_OFI_TCP,
+ CRT_PROV_OFI_TCP_RXM,
+ CRT_PROV_OFI_CXI,
+ CRT_PROV_OFI_OPX,
+ CRT_PROV_OFI_LAST = CRT_PROV_OFI_OPX,
+ CRT_PROV_UCX_RC,
+ CRT_PROV_UCX_UD,
+ CRT_PROV_UCX_RC_UD,
+ CRT_PROV_UCX_RC_O,
+ CRT_PROV_UCX_UD_O,
+ CRT_PROV_UCX_RC_UD_O,
+ CRT_PROV_UCX_RC_X,
+ CRT_PROV_UCX_UD_X,
+ CRT_PROV_UCX_RC_UD_X,
+ CRT_PROV_UCX_DC_X,
+ CRT_PROV_UCX_TCP,
+ CRT_PROV_UCX_LAST = CRT_PROV_UCX_TCP,
+ /* Note: This entry should be the last valid one in enum */
+ CRT_PROV_COUNT,
+ CRT_PROV_UNKNOWN = -1,
+} crt_provider_t;
 
 crt_provider_t
 crt_prov_str_to_prov(const char *prov_str);

src/cart/crt_init.c

@@ -71,7 +71,8 @@ dump_envariables(void)
  "CRT_CTX_SHARE_ADDR", "CRT_CTX_NUM", "D_FI_CONFIG",
  "FI_UNIVERSE_SIZE", "CRT_ENABLE_MEM_PIN",
  "FI_OFI_RXM_USE_SRX", "D_LOG_FLUSH", "CRT_MRC_ENABLE",
- "CRT_SECONDARY_PROVIDER", "D_PROVIDER_AUTH_KEY", "D_PORT_AUTO_ADJUST"};
+ "CRT_SECONDARY_PROVIDER", "D_PROVIDER_AUTH_KEY", "D_PORT_AUTO_ADJUST",
+ "D_POLL_TIMEOUT"};
 
  D_INFO("-- ENVARS: --\n");
  for (i = 0; i < ARRAY_SIZE(envars); i++) {
@@ -545,17 +546,17 @@ prov_settings_apply(bool primary, crt_provider_t prov, crt_init_options_t *opt)
 
  if (prov == CRT_PROV_OFI_CXI)
  mrc_enable = 1;
- else {
- /* Use tagged messages for other providers, disable multi-recv */
- apply_if_not_set("NA_OFI_UNEXPECTED_TAG_MSG", "1");
- }
 
  d_getenv_int("CRT_MRC_ENABLE", &mrc_enable);
  if (mrc_enable == 0) {
  D_INFO("Disabling MR CACHE (FI_MR_CACHE_MAX_COUNT=0)\n");
  setenv("FI_MR_CACHE_MAX_COUNT", "0", 1);
  }
 
+ /* Use tagged messages for other providers, disable multi-recv */
+ if (prov != CRT_PROV_OFI_CXI && prov != CRT_PROV_OFI_TCP)
+ apply_if_not_set("NA_OFI_UNEXPECTED_TAG_MSG", "1");
+
  g_prov_settings_applied[prov] = true;
 }
 

src/cart/utils/crt_utils.c

@@ -134,18 +134,19 @@ crtu_progress_fn(void *data)
  while (opts.shutdown == 0)
  crt_progress(*p_ctx, 1000);
 
- if (opts.is_swim_enabled && idx == 0)
- crt_swim_disable_all();
+ if (opts.is_server) {
+ if (opts.is_swim_enabled && idx == 0)
+ crt_swim_disable_all();
 
- rc = crtu_drain_queue(*p_ctx);
- D_ASSERTF(rc == 0, "crtu_drain_queue() failed with rc=%d\n", rc);
+  rc = crtu_drain_queue(*p_ctx);
+  D_ASSERTF(rc == 0, "crtu_drain_queue() failed with rc=%d\n", rc);
 
- if (opts.delay_shutdown_sec > 0)
- sleep(opts.delay_shutdown_sec);
+ if (opts.delay_shutdown_sec > 0)
+ sleep(opts.delay_shutdown_sec);
+ }
 
  rc = crt_context_destroy(*p_ctx, 1);
- D_ASSERTF(rc == 0, "Failed to destroy context %p rc=%d\n",
- p_ctx, rc);
+ D_ASSERTF(rc == 0, "Failed to destroy context %p rc=%d\n", p_ctx, rc);
 
  pthread_exit(rc ? *p_ctx : NULL);
 

src/client/api/event.c

@@ -20,6 +20,12 @@
 static __thread daos_event_t ev_thpriv;
 static __thread bool ev_thpriv_is_init;
 
+/**
+ * Global progress timeout for synchronous operation
+ * busy-polling by default (0), timeout in us otherwise
+ */
+static uint32_t ev_prog_timeout;
+
 #define EQ_WITH_CRT
 
 #if !defined(EQ_WITH_CRT)
@@ -91,6 +97,8 @@ daos_eq_lib_init()
 
  eq_ref = 1;
 
+ d_getenv_int("D_POLL_TIMEOUT", &ev_prog_timeout);
+
 unlock:
  D_MUTEX_UNLOCK(&daos_eq_lock);
  return rc;
@@ -1261,7 +1269,7 @@ daos_event_priv_wait()
 
  /* Wait on the event to complete */
  while (evx->evx_status != DAOS_EVS_READY) {
- rc = crt_progress_cond(evx->evx_ctx, 0, ev_progress_cb, &epa);
+ rc = crt_progress_cond(evx->evx_ctx, ev_prog_timeout, ev_progress_cb, &epa);
 
  /** progress succeeded, loop can exit if event completed */
  if (rc == 0) {

src/client/array/dc_array.c

@@ -1822,7 +1822,7 @@ struct key_query_props {
  char akey_val;
  daos_recx_t recx;
  daos_size_t *size;
- daos_size_t max_epoch;
+ daos_epoch_t max_epoch;
  tse_task_t *ptask;
 };
 
@@ -1831,8 +1831,7 @@ free_query_cb(tse_task_t *task, void *data)
 {
  struct key_query_props *props = *((struct key_query_props **)data);
 
- if (props->array)
- array_decref(props->array);
+ array_decref(props->array);
  D_FREE(props);
  return 0;
 }
@@ -1905,7 +1904,7 @@ dc_array_get_size(tse_task_t *task)
  query_args->dkey = &kqp->dkey;
  query_args->akey = &kqp->akey;
  query_args->recx = &kqp->recx;
- query_args->max_epoch = NULL;
+ query_args->max_epoch = &kqp->max_epoch;
 
  rc = tse_task_register_comp_cb(task, free_query_cb, &kqp, sizeof(kqp));
  if (rc != 0)
@@ -2285,7 +2284,7 @@ add_record(daos_handle_t oh, daos_handle_t th, struct set_size_props *props, d_l
  daos_key_t *dkey;
  struct io_params *params = NULL;
  tse_task_t *io_task;
- bool free_iod_recxs = true;
+ bool free_params = true;
  int rc;
 
  D_ALLOC_PTR(params);
@@ -2336,7 +2335,7 @@ add_record(daos_handle_t oh, daos_handle_t th, struct set_size_props *props, d_l
  rc = tse_task_register_comp_cb(io_task, free_io_params_cb, &params, sizeof(params));
  if (rc)
  D_GOTO(err_task, rc);
- free_iod_recxs = false;
+ free_params = false;
 
  rc = tse_task_register_deps(props->ptask, 1, &io_task);
  if (rc)
@@ -2351,9 +2350,10 @@ add_record(daos_handle_t oh, daos_handle_t th, struct set_size_props *props, d_l
  if (io_task)
  tse_task_complete(io_task, rc);
 err:
- D_FREE(params);
- if (free_iod_recxs)
+ if (free_params) {
  D_FREE(iod->iod_recxs);
+ D_FREE(params);
+ }
  return rc;
 }
 
@@ -2478,6 +2478,7 @@ dc_array_set_size(tse_task_t *task)
  daos_obj_list_dkey_t *enum_args;
  struct set_size_props *set_size_props = NULL;
  tse_task_t *enum_task;
+ bool cleanup = true;
  int rc;
 
  args = daos_task_get_args(task);
@@ -2537,31 +2538,34 @@ dc_array_set_size(tse_task_t *task)
  enum_args->sgl = &set_size_props->sgl;
  enum_args->dkey_anchor = &set_size_props->anchor;
 
- rc = tse_task_register_cbs(enum_task, NULL, NULL, 0, adjust_array_size_cb, &set_size_props,
- sizeof(set_size_props));
+ rc = tse_task_register_comp_cb(task, free_set_size_cb, &set_size_props,
+  sizeof(set_size_props));
  if (rc)
  D_GOTO(err_enum_task, rc);
+ cleanup = false;
 
- rc = tse_task_register_deps(task, 1, &enum_task);
+ rc = tse_task_register_comp_cb(enum_task, adjust_array_size_cb, &set_size_props,
+ sizeof(set_size_props));
  if (rc)
  D_GOTO(err_enum_task, rc);
 
- rc = tse_task_register_comp_cb(task, free_set_size_cb, &set_size_props,
- sizeof(set_size_props));
+ rc = tse_task_register_deps(task, 1, &enum_task);
  if (rc)
  D_GOTO(err_enum_task, rc);
 
  rc = tse_task_schedule(enum_task, true);
  if (rc)
- D_GOTO(err_enum_task, rc);
+ D_GOTO(err_task, rc);
 
  return 0;
 err_enum_task:
  tse_task_complete(enum_task, rc);
 err_task:
- D_FREE(set_size_props);
- if (array)
- array_decref(array);
  tse_task_complete(task, rc);
+ if (cleanup) {
+ D_FREE(set_size_props);
+ if (array)
+ array_decref(array);
+ }
  return rc;
 } /* end daos_array_set_size */