Skip to content
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

DAOS-16809 vos: container based stable epoch #15605

Merged
merged 3 commits into from
Jan 17, 2025
Merged

DAOS-16809 vos: container based stable epoch #15605

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
7aa0703
DAOS-16809 vos: container based stable epoch
Nasf-Fan Dec 12, 2024
1245805
DAOS-16809 engine: fix for review feedback
Nasf-Fan Jan 15, 2025
021833f
Merge branch 'master' into Nasf-Fan/DAOS-16809_1
Nasf-Fan Jan 17, 2025
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
9 changes: 8 additions & 1 deletion src/container/srv_target.c
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,6 +1,7 @@
/**
* (C) Copyright 2016-2024 Intel Corporation.
* (C) Copyright 2025 Google LLC
* (C) Copyright 2025 Hewlett Packard Enterprise Development LP
*
* SPDX-License-Identifier: BSD-2-Clause-Patent
*/
Expand Down Expand Up @@ -322,7 +323,7 @@ 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 = d_sec2hlc(vos_get_agg_gap());

D_ASSERT(hlc > (interval * 2));
/*
Expand Down Expand Up @@ -410,6 +411,9 @@ 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)
Expand All @@ -426,6 +430,9 @@ 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);
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Why do we need to set this value here and there? Isn't it a global constant value 'cur_time - gap'? I don't see why it's related to aggregation range.

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

If aggregation run slowly, much behind of current HLC, it is unnecessary to restart the DTX which epoch is out of 'cur_time - gap' but newer than current aggregation boundary. We only need to guarantee that the DTX's epoch is not smaller than current aggregation real boundary.

Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I see, thanks. but I don't think this optimization is necessary though. :)


if (dss_xstream_is_busy())
flags &= ~VOS_AGG_FL_FORCE_MERGE;
rc = agg_cb(cont, &epoch_range, flags, param);
Expand Down
2 changes: 2 additions & 0 deletions src/dtx/dtx_common.c
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,5 +1,6 @@
/**
* (C) Copyright 2019-2024 Intel Corporation.
* (C) Copyright 2025 Hewlett Packard Enterprise Development LP
*
* SPDX-License-Identifier: BSD-2-Clause-Patent
*/
Expand Down Expand Up @@ -922,6 +923,7 @@ 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;
Expand Down
4 changes: 3 additions & 1 deletion src/dtx/tests/dts_structs.c
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,5 +1,6 @@
/**
* (C) Copyright 2024 Intel Corporation.
* (C) Copyright 2025 Hewlett Packard Enterprise Development LP
*
* SPDX-License-Identifier: BSD-2-Clause-Patent
*/
Expand Down Expand Up @@ -70,8 +71,9 @@ 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);
Expand Down
26 changes: 13 additions & 13 deletions src/engine/sched.c
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,5 +1,6 @@
/**
* (C) Copyright 2016-2024 Intel Corporation.
* (C) Copyright 2025 Hewlett Packard Enterprise Development LP
*
* SPDX-License-Identifier: BSD-2-Clause-Patent
*/
Expand Down Expand Up @@ -197,17 +198,6 @@ 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 */
Expand Down Expand Up @@ -943,12 +933,22 @@ 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 against 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) <
(vos_get_agg_gap() + 10) * 1000; /* msecs */
}

static inline uint64_t
Expand Down
12 changes: 1 addition & 11 deletions src/include/daos/dtx.h
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,5 +1,6 @@
/**
* (C) Copyright 2019-2023 Intel Corporation.
* (C) Copyright 2025 Hewlett Packard Enterprise Development LP
*
* SPDX-License-Identifier: BSD-2-Clause-Patent
*/
Expand Down Expand Up @@ -27,17 +28,6 @@
/* 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),
Expand Down
7 changes: 6 additions & 1 deletion src/include/daos_srv/dtx_srv.h
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,5 +1,6 @@
/**
* (C) Copyright 2019-2024 Intel Corporation.
* (C) Copyright 2025 Hewlett Packard Enterprise Development LP
*
* SPDX-License-Identifier: BSD-2-Clause-Patent
*/
Expand Down Expand Up @@ -113,8 +114,10 @@ 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;
Expand Down Expand Up @@ -287,6 +290,8 @@ enum dtx_flags {
DTX_RELAY = (1 << 10),
/** Local transaction */
DTX_LOCAL = (1 << 11),
/** Locally generate the epoch. */
DTX_EPOCH_OWNER = (1 << 12),
};

void
Expand Down
51 changes: 51 additions & 0 deletions src/include/daos_srv/vos.h
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,5 +1,6 @@
/**
* (C) Copyright 2015-2024 Intel Corporation.
* (C) Copyright 2025 Hewlett Packard Enterprise Development LP
*
* SPDX-License-Identifier: BSD-2-Clause-Patent
*/
Expand Down Expand Up @@ -939,6 +940,56 @@ 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.
*
Expand Down
3 changes: 3 additions & 0 deletions src/include/daos_srv/vos_types.h
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,5 +1,6 @@
/**
* (C) Copyright 2015-2024 Intel Corporation.
* (C) Copyright 2025 Hewlett Packard Enterprise Development LP
*
* SPDX-License-Identifier: BSD-2-Clause-Patent
*/
Expand Down Expand Up @@ -58,6 +59,8 @@ 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 {
Expand Down
14 changes: 11 additions & 3 deletions src/object/srv_obj.c
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,5 +1,6 @@
/**
* (C) Copyright 2016-2024 Intel Corporation.
* (C) Copyright 2025 Hewlett Packard Enterprise Development LP
*
* SPDX-License-Identifier: BSD-2-Clause-Patent
*/
Expand Down Expand Up @@ -2898,8 +2899,10 @@ 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;
Expand Down Expand Up @@ -3858,8 +3861,10 @@ 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;
Expand Down Expand Up @@ -5112,6 +5117,7 @@ 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
Expand Down Expand Up @@ -5703,8 +5709,10 @@ 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);
Expand Down
2 changes: 2 additions & 0 deletions src/tests/ftest/util/server_utils_params.py
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,5 +1,6 @@
"""
(C) Copyright 2020-2024 Intel Corporation.
(C) Copyright 2025 Hewlett Packard Enterprise Development LP

SPDX-License-Identifier: BSD-2-Clause-Patent
"""
Expand Down Expand Up @@ -441,6 +442,7 @@ class EngineYamlParameters(YamlParameters):
"D_LOG_FILE_APPEND_PID=1",
"DAOS_POOL_RF=4",
"CRT_EVENT_DELAY=1",
"DAOS_VOS_AGG_GAP=25",
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

  1. Why change this from the default of 60?
  2. I would think we need to run this PR with affected tests? Maybe Features: aggregation?

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

  1. Why change this from the default of 60?
  2. I would think we need to run this PR with affected tests? Maybe Features: aggregation?

The environment variable DAOS_VOS_AGG_GAP is new introduced via this patch. Before that, the gap between aggregation and current HLC was DAOS_AGG_THRESHOLD, that was about 20 seconds. Such gap cannot match the new requirement for stable epoch in this patch because it was fixed and may be too short as to cause some transactions to be restarted frequently.
Generally, a small nvme device is more easy to be filled, so need more small aggregation gap, that is usually for test environment, such as for CI test. But in real product environment, we can use relative large configuration, that is why we say the fault is 60 seconds.

VOS aggregation will automatically run at background during CI tests except it is disabled explicitly.

Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

My main concern is the default pr tests might not be sufficient to cover this change. Even though aggregation is enabled by default, most tests do not explicitly verify aggregation behavior. I don't know this code well enough to say whether or not pr is enough.
I don't want to block this PR unnecessarily so I will remove my -1, but please keep in mind that in general it is much more expensive to fix regressions after initial landing than in the original PR, where all we have to do is use Features: <tags>

Copy link
Contributor

@mchaarawi mchaarawi Dec 18, 2024 •
edited
Loading

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I have some concerns about setting this flag for all tests. CI testing should resemble real world testing on functional HW. the per SSD capacity in CI functional tests is not different from production systems, where also the capacity varies greatly depending on what customers want.
It's fine to change it for some particular test, but (correct me if im wrong) this change here is for all CI tests?
If that is the case, we should not land that TBH unless there is some agreement that the default should change.

Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Yeah, this is for all functional CI tests

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

"DAOS_VOS_AGG_GAP=25" this setting makes CI tests almost keep the same behavior as without the patch (was 20+ seconds before). If without such setting, then for some short-time test, VOS aggregation may not be triggered before the test completed, that will decrease the potential race windows with VOS aggregation. Then may hide potential bugs. We can remove such setting for CI tests, but I am afraid that:

  1. Space pressure for small pool configuration.
  2. Less race with VOS aggregation.

Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Shouldn't we use this value (used by CI) as default and tune the value on larger system when necessary?

Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

It looks to me not quite right if we don't test default configuration in CI hardware tests. If the default value won't work well on extreme large system like Aurora, I think it's acceptable to ask user to configure a larger value (via server yaml).

"COVFILE=/tmp/test.cov"],
"ofi+tcp": [],
"ofi+tcp;ofi_rxm": [],
Expand Down
7 changes: 5 additions & 2 deletions src/utils/ddb/tests/ddb_test_driver.c
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,5 +1,6 @@
/**
* (C) Copyright 2022-2024 Intel Corporation.
* (C) Copyright 2025 Hewlett Packard Enterprise Development LP
*
* SPDX-License-Identifier: BSD-2-Clause-Patent
*/
Expand Down Expand Up @@ -410,6 +411,7 @@ dvt_dtx_begin_helper(daos_handle_t coh, const daos_unit_oid_t *oid, daos_epoch_t
struct dtx_handle *dth;
struct dtx_memberships *mbs;
size_t size;
int rc;

D_ALLOC_PTR(dth);
assert_non_null(dth);
Expand Down Expand Up @@ -449,7 +451,8 @@ dvt_dtx_begin_helper(daos_handle_t coh, const daos_unit_oid_t *oid, daos_epoch_t
dth->dth_shares_inited = 1;

vos_dtx_rsrvd_init(dth);
vos_dtx_attach(dth, false, false);
rc = vos_dtx_attach(dth, false, false);
assert_rc_equal(rc, 0);

*dthp = dth;
}
Expand Down Expand Up @@ -478,7 +481,7 @@ dvt_vos_insert_dtx_records(daos_handle_t coh, uint32_t nr, uint32_t committed_nr
daos_recx_t recxs[recxs_nr];
daos_iod_t iod = {0};
d_sg_list_t sgl = {0};
daos_epoch_t epoch = 1;
daos_epoch_t epoch = d_hlc_get();
uint64_t dkey_hash = 0x123;
int i;

Expand Down
6 changes: 5 additions & 1 deletion src/vos/tests/vts_dtx.c
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,5 +1,6 @@
/**
* (C) Copyright 2019-2024 Intel Corporation.
* (C) Copyright 2025 Hewlett Packard Enterprise Development LP
*
* SPDX-License-Identifier: BSD-2-Clause-Patent
*/
Expand Down Expand Up @@ -43,6 +44,7 @@ vts_dtx_begin(const daos_unit_oid_t *oid, daos_handle_t coh, daos_epoch_t epoch,
uint64_t dkey_hash, struct dtx_handle **dthp)
{
struct dtx_handle *dth;
int rc;

D_ALLOC_PTR(dth);
assert_non_null(dth);
Expand All @@ -66,6 +68,7 @@ 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;
Expand All @@ -91,7 +94,8 @@ vts_dtx_begin(const daos_unit_oid_t *oid, daos_handle_t coh, daos_epoch_t epoch,
dth->dth_shares_inited = 1;

vos_dtx_rsrvd_init(dth);
vos_dtx_attach(dth, false, false);
rc = vos_dtx_attach(dth, false, false);
assert_rc_equal(rc, 0);

*dthp = dth;
}
Expand Down
Loading
Loading