Skip to content

Merge release/2.6 into google/2.6 - #17135

Open
wangdi1 wants to merge 44 commits into
google/2.6from
ddiwang/google/2.6
Open

Merge release/2.6 into google/2.6#17135
wangdi1 wants to merge 44 commits into
google/2.6from
ddiwang/google/2.6

Conversation

@wangdi1

@wangdi1 wangdi1 commented Nov 16, 2025

Copy link
Copy Markdown
Contributor

liuxuezhao and others added 30 commits October 15, 2025 15:43
@liuxuezhao
DAOS-17589 pool: not changing fseq for exclude during drain (#16753)
3585d8d 
If the target is being drained, then exclude should not change
its fseq number, otherwise it may get wrong object layout and
interfere the drain process.

Signed-off-by: Di Wang <ddiwang@google.com>
Co-authored-by: Xuezhao Liu <xuezhao.liu@hpe.com>
@daltonbohning
DAOS-18029 cq: do not run trivy on all prs (#16935) (#16957)
b9c6229 
Trivy often fails and is unrelated to most PRs. Only run it on PRs if
relevant files are touched.

Signed-off-by: Dalton Bohning <dalton.bohning@hpe.com>
@wangshilong @mchaarawi
DAOS-18103: container: fix compatibility for newer clients with older…
599f078 
… servers (#16981)

Now the pool connection will return the server's maximum supported layout version,
allowing the client to determine which version to use for container creation/open.

Signed-off-by: Wang Shilong <shilong.wang@hpe.com>
Co-authored-by: Mohamad Chaarawi <mohamad.chaarawi@hpe.com>
@gnailzenh
DAOS-17772 rebuild: avoid potential races between rebuild and EC aggr…
c979f32 
…egation (#16945)

Increment the map_version for reclaim tasks to ensure they operate on the latest
pool map. Without this, reclaim tasks might act on an outdated pool map, potentially
causing data corruption.

Always recalculate parity for partial overwrites, disabling the optimized path to
ensure data consistency.

Ignore parity update errors during EC aggregation. This is a temporary workaround
because DAOS does not use DTX to update parities during EC aggregation.
Without DTX, parity shards may become inconsistent if an engine failure occurs
alongside a rebuild. This approach significantly increases the likelihood of parity
shards remaining consistent when EC aggregation interacts with rebuild operations.

Standardize the parity target status checks for both full stripe and hole processing
during EC aggregation.

Introduce additional barriers to minimize the chances of overlapping EC aggregation
and rebuild operations.

Signed-off-by: Liang Zhen <gnailzenh@gmail.com>
@phender
DAOS-18050 build: Tag 2.6.4 rc3 (#16994)
e854411 
Tag third release candidate for 2.6.4.

Signed-off-by: Phil Henderson <phillip.henderson@hpe.com>
@liuxuezhao
DAOS-18114 container: add mutex to protect cs_ec_agg_list (#17000)
5fde7ff 
Signed-off-by: Xuezhao Liu <xuezhao.liu@hpe.com>
@liuxuezhao
DAOS-18110 test: refine RF 2 test cases (#17008)
35c1183 
remove the inflight update test to avoid DER_UPDATE_AGAIN trouble,
it is not easy to control the timing for that case.

Signed-off-by: Xuezhao Liu <xuezhao.liu@hpe.com>
@wangshilong
DAOS-18127 container: use VERSION 1 for backward compatibility in con…
06acdf4 
…tainer creation (#17018)

New container creation by clients will specify the object version.
If not specified (dpe_val == 0), it indicates a client from before
DAOS 2.6.4, so use VERSION 1 for backward compatibility instead
of inheriting from pool object version.

Signed-off-by: Wang Shilong <shilong.wang@hpe.com>
@ryon-jensen
DAOS-18121 build: Update dependency versions (#17006)
0d3f0bf 
- Bump libfabric version >= 1.22.0-4
- Bump mercury version >= 2.4.0-7
- Bump argobots version >= 1.2-3
- Bump libisal version >= 2.31.1-7
- Bump libisal_crypto version >= 2.24.0-3
- Bump pmemobj1 >= 2.1.0-3
- Bump pmemobj >= 2.1.0-6

Signed-off-by: Ryon Jensen <ryon.jensen@hpe.com>
@phender
DAOS-18129 build: Tag 2.6.4 rc4 (#17024)
70bcf19 
Tag fourth release candidate for 2.6.4

Signed-off-by: Phil Henderson <phillip.henderson@hpe.com>
@gnailzenh @mchaarawi
DAOS-9355 doc: DAOS 2.6.4 release notes (#17047)
937eafa 
Signed-off-by: Liang Zhen <gnailzenh@gmail.com>
Co-authored-by: Mohamad Chaarawi <mohamad.chaarawi@hpe.com>
@mchaarawi
DAOS-18121 build: Update libfabric dep version (#17066)
b7dfc66 
revert libfabric dep version to 1.20 to account for Slingshot libfabric
on aurora.

Signed-off-by: Mohamad Chaarawi <mohamad.chaarawi@hpe.com>
@mchaarawi
DAOS-18129 build: update the ubuntu changelog and release notes (#17073)
4ab491c 
Signed-off-by: Mohamad Chaarawi <mohamad.chaarawi@hpe.com>
@phender
DAOS-18129 build: Tag 2.6.4 rc5 (#17074)
06de8a0 
Tag fifth release candidate for 2.6.4

Signed-off-by: Phil Henderson <phillip.henderson@hpe.com>
@liw
DAOS-17442 engine: Improve start issue handling (#16376) (#16588)
ef5e69f 
We infer from the available log messages that bio_xsctxt_alloc might get stuck after "failed to init bdevs: -1003" for some unknown reason related to SPDK. This patch skips two while statements in bio_xsctxt_free that might be where we got stuck.

When not in the CR mode, a pool ignores nonexistent children, leading to states that the current code cannot handle well (yet). This patch restricts this CR behavior to the CR mode. (Ideally, we shall allow the healthy children to start and automatically exclude the nonexistent children in the future.)

To make pool start failures easier to spot, this patch adds a RAS event that is raised when we fail to start a pool.

Also, this patch

fixes "-1" xstream IDs in messages logged during engine setup,
fixes a few places where we should write "pool" instead of "pool service", including the name of an internal environment variable, and
adds some infrequent log messages to help future debugging of engine start issues.

Move the unusually late initialization of dss_tls_key and
daos_srv_modkey earlier, so that the TLS checks in server_id_cb will be
safe. Both objects are static, and there's no reason to initialize them
so late.

Signed-off-by: Li Wei <liwei@hpe.com>
@mchaarawi
DAOS-18171 dfs: set dfs chunk size for RF3 containers (#17060)
50bac88 
EC with 3 parity is supported, so we should consider that for the dfs
chunk size selection.

Signed-off-by: Mohamad Chaarawi <mohamad.chaarawi@hpe.com>
@dinghwah
DAOS-17911 test: fix dfuse/sparse_file.py (#16759) (#17078)
db80417 
Resolve for paramiko ssh issue

Signed-off-by: Ding-Hwa Ho <ding-hwa.ho@hpe.com>
@Nasf-Fan
DAOS-17258 object: more check for CPD RPC bulk transfer - b26 (#16082)
2c19a8b 
To detect potential on-wire data corruption for CPD RPC bulk transfer.
CPD RPC bulk logic improvement (use bulk handle instead of bulk handle
pointer on-wire) and cleanup.

Signed-off-by: Fan Yong <fan.yong@hpe.com>
@daltonbohning
DAOS-17998 test: enhance test_checker_on_admin_excluded (#16894) (#17106
e65a4d7 
)

- Verify ranks can start after clearing exclusions
- Stop a random rank instead of hardcoded
- Increase to 4 servers / dual engine

Signed-off-by: Dalton Bohning <dalton.bohning@hpe.com>
@daltonbohning
DAOS-18102 test: update agent_failure.py to run commands as user (#16980
ebdc6cc 
) (#17105)

Update get_journalctl to accept user argument other than root.
Update stop_processes to accept user argument other than root.
Update agent_failure.py to run these as the current user when
applicable.

Signed-off-by: Dalton Bohning <dalton.bohning@hpe.com>
@daltonbohning
DAOS-17461 cq: speedup githook submodule check (#16306) (#16442)
7f9812c 
Re-write the check so it does not re-run the same subprocess commands.

Signed-off-by: Dalton Bohning <dalton.bohning@hpe.com>
@daltonbohning
DAOS-17406 cq: use cancel-in-progress for more workflows (#16244) (#1…
b54fe5e 
…6266)

So previous runs are cancelled on new pushes.

Signed-off-by: Dalton Bohning <dalton.bohning@hpe.com>
@liw @tanabarr
DAOS-17659 control: Enable setting property value with multiple strin…
5920e47 
…gs (#16503) (#16703)

Fix issue where a property value containing multiple comma-separated strings cannot be set e.g. self_heal:exclude,rebuild.

New syntax for key-value positional argument:
key:val[,key:val1[;val2...]...]
e.g. dmg pool set-prop bob self_heal:"exclude;rebuild"

Also includes the fix for DAOS-17666.

Signed-off-by: Tom Nabarro <thomas.nabarro@hpe.com>
Signed-off-by: Li Wei <liwei@hpe.com>
Co-authored-by: Tom Nabarro <tom.nabarro@intel.com>
@NiuYawei
DAOS-17628 bio: flush WAL header before unmap (#16478) (#16663)
08d5909 
In bio_wal_checkpoint(), we shouldn't unmap the checkpoint-ed region
before flushing WAL header (to make the last checkpoint-ed ID
persistent), otherwise, if the engine is interrupted in between the
unmap and flush, next WAL replay on engine start will replay from
the stale checkpoint-ed ID where the WAL tx data is already cleared
by unmap.

Signed-off-by: Niu Yawei <yawei.niu@hpe.com>
@grom72
DAOS-18065 ci: empty /sys/class/nvme/ on leap15 when no NVMe device a…
924eba4 
…ttached (#16956) (#17013)

On leap15 the /sys/class/nvme/ folder exists but it is empty
when no NVMe device attached.

Signed-off-by: Tomasz Gromadzki <tomasz.gromadzki@hpe.com>
@daltonbohning
DAOS-17429 cq: use merge-base in clang-format (#16269) (#17102)
c0cb06f 
Use the merge-base in the GHA clang-format instead of the tip of the target

Signed-off-by: Dalton Bohning <dalton.bohning@hpe.com>
@Nasf-Fan
DAOS-17977 vos: new parameter acts for sys_db_trav_cb_t - b26 (#16960)
fd3bd1e 
db_traverse() calls vos_iterate() to traverse the VOS_ITER_AKEY
for related sysdb table. For each found item, it handles it via
sys_db_trav_cb_t() that may remove such item from related sysdb
table, such as for some CHK cleanup use case. If that happened,
We will notify vos_iterate() logic via set "VOS_ITER_CB_DELETE"
flag in the "acts" parameter; otherwise some item(s) in related
sysdb table may be skipped in subsequent iteration. For CHK use
case, that may affect new check instance to reset automatically.

Signed-off-by: Fan Yong <fan.yong@hpe.com>
@Nasf-Fan
DAOS-17822 chk: filter out repeated pools in the check list - b26 (#1…
3e541c9 
…6625)

The user/admin may specify repeated pool(s) in the check list via
'dmg check' command. The logic needs to handle such case to avoid
invalid assertion.

The patch also fixes chk dump logic to avoid out-of-boundary DRAM
accessing.

Signed-off-by: Fan Yong <fan.yong@hpe.com>
@daltonbohning
DAOS-18101 test: support DAOS_TEST_SYSTEMD_PATH and LIBRARY_PATH (#16979
fede8fd 
) (#17104)

Support passing DAOS_TEST_SYSTEMD_PATH and DAOS_TEST_SYSTEMD_LIBRARY_PATH
to ftest. When set, this tells ftest to put these values into the
systemd config for the server and agent.

Signed-off-by: Dalton Bohning <dalton.bohning@hpe.com>
@daltonbohning
DAOS-17890 test: condense rebuild/basic.py (#16736) (#16816)
e296d9e 
Combine all the variants into one complete variant.
This is closer to production and is more efficient.

Signed-off-by: Dalton Bohning <dalton.bohning@hpe.com>
kjacque and others added 14 commits November 12, 2025 12:44
@kjacque
DAOS-16565 control: Log component check at Trace level (#16329) (#16806)
bd7c85b 
The component compatibility check is noisy, and reduces the
usefulness of debug-level logging for customers. It's still
useful for devs as a trace-level message.

Signed-off-by: Kris Jacque <kris.jacque@hpe.com>
@kjacque
DAOS-17643 control: Disallow most transitions from Excluded (#16475) (#…
8a20f44 
…17093)

Excluded ranks should be able to move only to the Joined or
AdminExcluded states. Regardless of the process state, without
re-joining successfully, the rank is still effectively excluded.

We saw a real life case where Excluded ranks were transitioning
to the Errored state when they SIGKILLed themselves after
discovering their exclusion.

Signed-off-by: Kris Jacque <kris.jacque@hpe.com>
@phender
SRE-3150 test: Add device verification and test summary to launch (#1…
eda8d6b 
…6583) (#16991)

Verify the server storage devices exist before and after running the
tests.

Generate a test summary indicating the order, status, and host usage for
each test file and its variants.

Signed-off-by: Phil Henderson <phillip.henderson@hpe.com>
@Nasf-Fan
DAOS-17845 container: keep CONT_TGT_SNAPSHOT_NOTIFY RPC errno - b26 (#…
a42f21d 
…16673)

Return CONT_TGT_SNAPSHOT_NOTIFY RPC errno to the snap_oit_create()
sponsor instead of converting it as -DER_IO. Then related client
side logic can know what the failure is.

Signed-off-by: Fan Yong <fan.yong@hpe.com>
@frostedcmos
DAOS-17871 cart: Improve swim notification test (#16752) (#17119)
d8c1016 
Change how the test behaves. At start it will now wait for all 3 ranks
to be present and waits 30 seconds after that before rank=2 is shut down.

30 seconds is left to allow for swim notification status to be propagated across
all 3 servers in this group.

Added --skip_wait option to clients to skip crtu_wait_for_ranks() call as only
first client invocation needs to ensure all 3 servers are up.

Signed-off-by: Alexander A Oganezov <alexander.oganezov@hpe.com>
@frostedcmos
DAOS-17470 cart: add out of quota RPC to timeout (#16308) (#16582)
af379a2 
Add out of quota RPC to timeout list once resubmit them.

Signed-off-by: Di Wang <ddiwang@google.com>
Signed-off-by: Alexander A Oganezov <alexander.oganezov@hpe.com>
@knard38
DAOS-17286 gurt: fix rank set string convertion (#16204) (#16591)
6abffd7 
Remove string maximal length limitation of the d_rank_range_list_str()
conversion function.

Signed-off-by: Cedric Koch-Hofer <cedric.koch-hofer@hpe.com>
@janekmi @Nasf-Fan
DAOS-17508 vos: allow DAOS_INTENT_CHECK for EV iterators (#16737) (#1…
f784c5b 
…6784)

DAOS_INTENT_CHECK is intended for offline utilities like DDB and DLCK
to inspect the contents of the pool while it is idle.
EV iterators must support it as well.

Signed-off-by: Jan Michalski <jan-marian.michalski@hpe.com>
Co-authored-by: Fan Yong <fan.yong@hpe.com>
@kjacque @tanabarr
DAOS-17854 control: Only add voter if it doesn't exist (#16721) (#17117)
601bc29 
Don't modify the voter list unless the voter isn't already listed.

Signed-off-by: Kris Jacque <kris.jacque@hpe.com>
Co-authored-by: Tom Nabarro <thomas.nabarro@hpe.com>
@jgmoore-or
DAOS-16478 test: Change pool autotest to use DAOS_EQ_NOWAIT. (#15936) (…
3835e93 
…#17089)

Signed-off-by: Joseph Moore <joseph.moore@hpe.com>
@dependabot
DAOS-18211 cq: bump GHA versions (#17096) (#17097)
ebb845d 
Updates `EnricoMi/publish-unit-test-result-action` from 2.20.0 to 2.21.0
Updates `actions/upload-artifact` from 4 to 5
Updates `codespell-project/actions-codespell` from 2.1 to 2.2
Updates `github/codeql-action` from 4.30.7 to 4.31.2

Signed-off-by: dependabot[bot] <support@github.com>
Signed-off-by: Dalton Bohning <dalton.bohning@hpe.com>
@kjacque
DAOS-17865 control: Prevent duplicate pool props (#16883) (#17116)
4373168 
Due to a bug in parsing props, it was possible to crash the daos_engine by entering the same pool property twice during pool create by using the --properties flag twice. A prop could also be duplicated within the --properties flag without causing an error.

This patch checks for duplication at three different points:
- Within all props set within a --properties flag.
- Within different property strings, if --properties was used multiple times.
- At the engine level.

Signed-off-by: Kris Jacque <kris.jacque@hpe.com>
@wangdi1
Merge remote-tracking branch 'origin/release/2.6' into ddiwang/google…
937e5ef 
…/2.6

Signed-off-by: Di Wang <ddiwang@google.com>
@wangdi1
Merge from release-2.6
1bb3538 
Signed-off-by: Di Wang <ddiwang@google.com>
@github-actions

github-actions Bot commented Nov 16, 2025

Copy link
Copy Markdown

Errors are component not formatted correctly,Ticket number prefix incorrect,PR title is malformatted. See https://daosio.atlassian.net/wiki/spaces/DC/pages/11133911069/Commit+Comments,Unable to load ticket data
https://daosio.atlassian.net/browse/Merge

@daosbuild3

daosbuild3 commented Nov 16, 2025

Copy link
Copy Markdown
Collaborator

Test stage Functional on EL 8.8 completed with status FAILURE. https://jenkins-3.daos.hpc.amslabs.hpecorp.net//job/daos-stack/job/daos/view/change-requests/job/PR-17135/1/execution/node/1133/log

This branch has not been deployed

No deployments
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

@jxiong jxiong jxiong approved these changes
@jolivier23 jolivier23 Awaiting requested review from jolivier23
@mjmac mjmac Awaiting requested review from mjmac

Assignees

No one assigned

Labels

None yet

Milestone

No milestone

Development

Successfully merging this pull request may close these issues.