Skip to content

DAOS-19591 container: re-send EC agg boundary to ranks waiting for it #19304

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
mchaarawi merged 1 commit into from
Oct 16, 2026
Merged

DAOS-19591 container: re-send EC agg boundary to ranks waiting for it #19304

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
61 changes: 34 additions & 27 deletions src/container/srv_container.c
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1778,6 +1778,22 @@ ds_cont_leader_update_agg_eph(uuid_t pool_uuid, uuid_t cont_uuid,
if (ec_agg->ea_server_ephs[i].eph < eph) {
ec_agg->ea_server_ephs[i].eph = eph;
ec_agg->ea_server_ephs[i].ee_update_ts = daos_gettime_coarse();
} else if (eph == 0 && ec_agg->ea_current_eph != 0) {
/*
* A rank reports zero only while its EC aggregation is paused
* waiting for the boundary (sc_ec_agg_eph_valid == 0), e.g. a
* reintegrated/extended rank that got the VOS container after the
* last refresh. Ask cont_agg_eph_sync() to re-send the current
* boundary even if the minimum epoch did not change, otherwise
* the rank never starts and pins the boundary forever.
* (ea_current_eph == 0 refreshes every round anyway.)
*/
if (!ec_agg->ea_need_refresh)
D_INFO(DF_CONT ": rank %u waits for EC agg boundary, "
"current " DF_X64 "\n",
DP_CONT(pool_uuid, cont_uuid), rank,
ec_agg->ea_current_eph);
ec_agg->ea_need_refresh = 1;
}
break;
}
Expand Down Expand Up @@ -1820,25 +1836,9 @@ cont_refresh_vos_agg_eph_one(void *data)
int rc;

rc = ds_cont_child_lookup(arg->pool_uuid, arg->cont_uuid, &cont_child);
if (rc == -DER_NONEXIST) {
/*
* The VOS container can be absent on this target when the container was
* created while the target was excluded, and the following reintegration
* (or extend) did not migrate any record for it. Create it here, otherwise
* this target never reports EC aggregation epoch to the container service
* leader, and the leader keeps the EC aggregation boundary of the whole
* container pinned. The container service leader only refreshes containers
* that still exist in RDB, so it is safe to create the VOS container here.
*/
rc = ds_cont_child_open_create(arg->pool_uuid, arg->cont_uuid, &cont_child);
DL_CDEBUG(rc != 0 && rc != -DER_SHUTDOWN, DLOG_ERR, DLOG_INFO, rc,
DF_CONT " create missing vos container",
DP_CONT(arg->pool_uuid, arg->cont_uuid));
if (rc != 0)
return rc;
} else if (rc != 0) {
DL_CDEBUG(rc != -DER_SHUTDOWN, DLOG_ERR, DB_MD, rc, DF_CONT " lookup cont failed",
DP_CONT(arg->pool_uuid, arg->cont_uuid));
if (rc) {
DL_CDEBUG(rc != 0 && rc != -DER_SHUTDOWN, DLOG_ERR, DB_MD, rc,
DF_CONT " lookup cont failed", DP_CONT(arg->pool_uuid, arg->cont_uuid));
return rc;
}

Expand Down Expand Up @@ -1868,13 +1868,9 @@ ds_cont_tgt_refresh_agg_eph(uuid_t pool_uuid, uuid_t cont_uuid,
uuid_copy(arg.cont_uuid, cont_uuid);
arg.min_eph = eph;

/*
* NB: cont_refresh_vos_agg_eph_one() may create the VOS container, which can yield
* and needs a deep stack, so a thread (ULT) collective must be used here.
*/
rc = ds_pool_thread_collective(
pool_uuid, PO_COMP_ST_NEW | PO_COMP_ST_DOWN | PO_COMP_ST_DOWNOUT,
cont_refresh_vos_agg_eph_one, &arg, DSS_ULT_DEEP_STACK | DSS_ULT_FL_PERIODIC);
rc = ds_pool_task_collective(pool_uuid,
PO_COMP_ST_NEW | PO_COMP_ST_DOWN | PO_COMP_ST_DOWNOUT,
cont_refresh_vos_agg_eph_one, &arg, DSS_ULT_FL_PERIODIC);
if (rc) {
DL_ERROR(rc, DF_CONT ": refresh ec_agg_eph " DF_X64 " failed.",
DP_CONT(pool_uuid, cont_uuid), eph);
Expand Down Expand Up @@ -2096,7 +2092,14 @@ cont_agg_eph_sync(struct ds_pool *pool, struct cont_svc *svc)
if (min_eph < ec_agg->ea_rdb_eph)
min_eph = ec_agg->ea_rdb_eph;

if (min_eph == ec_agg->ea_current_eph && ec_agg->ea_current_eph != 0)
/*
* Skip the refresh when nothing changed, unless some rank asked for the
* boundary (see ds_cont_leader_update_agg_eph()). Such a rank has eph 0,
* so min_eph is pinned to ea_rdb_eph == ea_current_eph and it would never
* get the boundary it needs to start reporting.
*/
if (min_eph == ec_agg->ea_current_eph && ec_agg->ea_current_eph != 0 &&
!ec_agg->ea_need_refresh)
continue;

/**
Expand Down Expand Up @@ -2143,6 +2146,10 @@ cont_agg_eph_sync(struct ds_pool *pool, struct cont_svc *svc)
}
}

/* clear before the refresh (which yields), so a request arriving meanwhile
* triggers another refresh in the next round.
*/
ec_agg->ea_need_refresh = 0;
rc = cont_iv_ec_agg_eph_refresh(pool->sp_iv_ns, ec_agg->ea_cont_uuid, min_eph,
svc->cs_ec_leader_ephs_req);
if (rc) {
Expand Down
6 changes: 5 additions & 1 deletion src/container/srv_internal.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -78,7 +78,11 @@ struct cont_ec_agg {
struct ec_eph *ea_server_ephs;
d_list_t ea_list;
int ea_servers_num;
uint32_t ea_deleted : 1;
uint32_t ea_deleted : 1,
/* some rank reported eph 0 (its EC aggregation waits for the boundary), re-send the
* boundary in the next cont_agg_eph_sync() even if the minimum epoch did not change.
*/
ea_need_refresh : 1;
};

/*
Expand Down
Loading