CL/HIER: team cleanup fix (#493)

openucx · May 4, 2022 · e96a6de · e96a6de
1 parent 246d4f4
commit e96a6de
Show file tree

Hide file tree

Showing 2 changed files with 13 additions and 10 deletions.
diff --git a/src/components/cl/hier/cl_hier_team.c b/src/components/cl/hier/cl_hier_team.c
@@ -52,12 +52,11 @@ UCC_CLASS_INIT_FUNC(ucc_cl_hier_team_t, ucc_base_context_t *cl_context,
 
  UCC_CLASS_CALL_SUPER_INIT(ucc_cl_team_t, &ctx->super, params);
 
+ memset(self->sbgps, 0, sizeof(self->sbgps));
  ucc_cl_hier_enable_sbgps(self);
  n_sbgp_teams = 0;
  for (i = 0; i < UCC_HIER_SBGP_LAST; i++) {
- hs = &self->sbgps[i];
- hs->score = NULL;
- hs->sbgp = NULL;
+ hs = &self->sbgps[i];
  if (hs->state == UCC_HIER_SBGP_ENABLED) {
  hs->sbgp = ucc_topo_get_sbgp(params->team->topo, hs->sbgp_type);
  if (hs->sbgp->status != UCC_SBGP_ENABLED) {
@@ -176,12 +175,16 @@ ucc_status_t ucc_cl_hier_team_destroy(ucc_base_team_t *cl_team)
  for (i = 0; i < UCC_HIER_SBGP_LAST; i++) {
  hs = &team->sbgps[i];
  if (hs->state == UCC_HIER_SBGP_ENABLED) {
- ucc_coll_score_free_map(hs->score_map);
+ if (hs->score_map) {
+ ucc_coll_score_free_map(hs->score_map);
+ }
  for (j = 0; j < hs->n_tls; j++) {
- ucc_tl_context_put(hs->tl_ctxs[j]);
- team->team_create_req
- ->descs[team->team_create_req->n_teams++]
- .team = hs->tl_teams[j];
+ if (hs->tl_teams[j]) {
+ ucc_tl_context_put(hs->tl_ctxs[j]);
+ team->team_create_req
+ ->descs[team->team_create_req->n_teams++]
+ .team = hs->tl_teams[j];
+ }
  }
  }
  }

diff --git a/src/core/ucc_team.c b/src/core/ucc_team.c
@@ -259,6 +259,7 @@ ucc_team_create_cls(ucc_context_t *context, ucc_team_t *team)
  if (status < 0) {
  team->n_cl_teams--;
  ucc_info("failed to create CL %s team", cl_iface->super.name);
+ cl_iface->team.destroy(b_team);
  } else if (status == UCC_INPROGRESS) {
  return status;
  }
@@ -270,13 +271,12 @@ ucc_team_create_cls(ucc_context_t *context, ucc_team_t *team)
  &team->bp, &b_team);
  if (status != UCC_OK) {
  ucc_info("failed to create CL %s team", cl_iface->super.name);
- /* TODO: see comment above*/
  continue;
  }
  status = cl_iface->team.create_test(b_team);
  if (status < 0) {
  ucc_info("failed to create CL %s team", cl_iface->super.name);
- /* TODO: see comment above */
+ cl_iface->team.destroy(b_team);
  continue;
  }
  team->cl_teams[team->n_cl_teams++] =