working around an unusual encounter when the joined_dim has actual value "max::size_t - 1" #1740

guj · 2025-03-28T00:58:52Z

from time to time, I see exception was caught for joined array cases here:

for (unsigned int i = 0; i < actualDim; i++)
{
if (!(joinedDim.has_value() && *joinedDim == i) &&
offset[i] + extent[i] > shape[I])
}
because offset is given as {} for joined array variable.

the shape of the joined var is 18446744073709551615

joined array:

from application:
- particle variable is initialized to openPMD::Dataset::JOINED_DIMENSION
definition in openPMD:
- Dataset.hpp: JOINED_DIMENSION = std::numeric_limitsstd::uint64_t::max()
- ADIOS2IOHandler.cpp: finding if (dims[i] == adios2::JoinedDim)
from ADIOS:
- ADIOSTypes.h: constexpr uint64_t JoinedDim = MaxU64 - 1;

The reason it reached here is because the joined dimension from ADIOS is 18446744073709551614, one less than the supposed max of size_t, 18446744073709551615 , which is the shape of the variable.

This pull request checked the variable shape ID and acknowledge as joined array if adios2::ShapeID::JoinedArary is returned.

franzpoeschel · 2025-03-28T11:44:04Z

include/openPMD/IO/ADIOS/ADIOS2IOHandler.hpp

-                        "[ADIOS2] Dataset access out of bounds.");
+                        "[ADIOS2] Offset must be an empty vector in case of "
+                        "joined "
+                        "array.");


This duplicates the condition checked in the if branch from line 492 to 509.
Can't we just change the if condition in line 492 to:

if (joinedDim.has_value() || var.ShapeID() == adios2::ShapeID::JoinedArray)

This would be a single-line change and do basically the same

It will break the for loop at 500

That issue only comes up when joinedDim.has_value() is false, but var.ShapeID() == adios2::ShapeID::JoinedArray is true, which we should treat as an error anyway?
I think I'll add a check for that circumstance.

guj

It will break the for loop at 500

guj · 2025-03-28T14:51:42Z

include/openPMD/IO/ADIOS/ADIOS2IOHandler.hpp

-                        "[ADIOS2] Dataset access out of bounds.");
+                        "[ADIOS2] Offset must be an empty vector in case of "
+                        "joined "
+                        "array.");


It will break the for loop at 500

franzpoeschel · 2025-04-01T11:53:05Z

I went through the implementation of joined arrays again and added some further safety guards. I think that I might have an idea on what's going on in your crashing simulations: Joined arrays were until now ignored in the extendDataset() task, meaning that openPMD::Dataset::JOINED_DIMENSION will be propagated as-is to the ADIOS2 variable. This would explain why you see ~(size_t)0 and not ~(size_t)0 - 1 (which would be adios2::JoinedDim.
However, I'm not sure ADIOS2 even supports extending joined array datasets. Also, since you are using joined arrays for particle output (which is 1D), there is nothing to be extended since the joined extent is calculated on the fly anyway.

franzpoeschel · 2025-04-01T15:58:17Z

@guj Can you try if you still observe the crashes with the recent commits on this PR? I hope that I might have found and fixed the issue.
I will not merge this yet, since I found another bug on the way and still have to find the cause for that.

franzpoeschel · 2025-04-01T18:23:57Z

I will not merge this yet, since I found another bug on the way and still have to find the cause for that.

Found and fixed the bug
Please still check if this fixes the issue you had, I think this might have been the problem

franzpoeschel · 2025-04-02T13:24:24Z

CI failures due to deployment of CMake 4.0 to CI runners, will be fixed with #1742. I'll select the bugfix for the double-write from unique-pointers into a separate PR as it has nothing to do with this PR. Will rebase this onto that then.

guj · 2025-04-02T16:15:09Z

@guj Can you try if you still observe the crashes with the recent commits on this PR? I hope that I might have found and fixed the issue. I will not merge this yet, since I found another bug on the way and still have to find the cause for that.

The test job was finally scheduled. Yes this version works!

guj · 2025-04-02T16:16:39Z

I will not merge this yet, since I found another bug on the way and still have to find the cause for that.

Found and fixed the bug Please still check if this fixes the issue you had, I think this might have been the problem

OK I will submit another test to double check.

franzpoeschel · 2025-04-02T16:31:06Z

The test job was finally scheduled. Yes this version works!

Nice, thanks for checking!

OK I will submit another test to double check.

I think it should not alter your use case since that's an entirely different issue, but feel free to run a test anyway

guj · 2025-04-03T05:26:05Z

I will not merge this yet, since I found another bug on the way and still have to find the cause for that.

Found and fixed the bug Please still check if this fixes the issue you had, I think this might have been the problem

OK I will submit another test to double check.

This one passed too

value "max::size_t - 1"

for more information, see https://pre-commit.ci

include/openPMD/IO/ADIOS/ADIOS2IOHandler.hpp

…lue "max::size_t - 1" (openPMD#1740) * working around an unusal encounter when the joined_dim has actual value "max::size_t - 1" * [pre-commit.ci] auto fixes from pre-commit.com hooks for more information, see https://pre-commit.ci * Add test for redundant resetDataset() * Merge check into above logic * Better error messages in verifyDataset * Add further safety guards to createDataset and extendDataset tasks * Move joinedDim logic into middle-end for extendDataset * Update include/openPMD/IO/ADIOS/ADIOS2IOHandler.hpp --------- Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com> Co-authored-by: Franz Pöschel <[email protected]>

* Fix: Late unique_ptr puts without CLOSE_FILE or ADVANCE operations (#1744) * Add failing test * Add failing test * Revert "Add failing test" This reverts commit 5e04ece. * Reactivate writing from unique_ptr in finalize() * BP5+groupbased: allow only up to 100 steps (#1732) * BP5+groupbased: allow only up to 1000 steps * Configure this via env variable OPENPMD_BP5_GROUPENCODING_MAX_STEPS=1000 * Add documentation * Lower limit to 100 * Add compile-time check for #1720 (#1722) * WarpX: Repo Moved (#1733) Update a link to WarpX. * Fix zero-sized storeChunk for Span API in Python (#1738) * working around an unusual encounter when the joined_dim has actual value "max::size_t - 1" (#1740) * working around an unusal encounter when the joined_dim has actual value "max::size_t - 1" * [pre-commit.ci] auto fixes from pre-commit.com hooks for more information, see https://pre-commit.ci * Add test for redundant resetDataset() * Merge check into above logic * Better error messages in verifyDataset * Add further safety guards to createDataset and extendDataset tasks * Move joinedDim logic into middle-end for extendDataset * Update include/openPMD/IO/ADIOS/ADIOS2IOHandler.hpp --------- Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com> Co-authored-by: Franz Pöschel <[email protected]> * ADIOS2 bugfix: Always use CurrentStep() in mode::Read (#1749) * Always use CurrentStep() in mode::Read * Remove manual step counting m_currentStep only necessary for SetStepSelection, it seems * Clean up logic that is no longer needed * Add test --------- Co-authored-by: Axel Huebl <[email protected]> Co-authored-by: Junmin Gu <[email protected]> Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>

franzpoeschel reviewed Mar 28, 2025

View reviewed changes

guj mentioned this pull request Mar 28, 2025

Joined Array support is added for ADIOS when writing out particles. BLAST-WarpX/warpx#5804

Open

1 task

guj commented Mar 31, 2025

View reviewed changes

guj changed the title ~~working around an unusal encounter when the joined_dim has actual value "max::size_t - 1"~~ working around an unusual encounter when the joined_dim has actual value "max::size_t - 1" Mar 31, 2025

franzpoeschel force-pushed the 20250328-joinedarray branch 2 times, most recently from b4be4a4 to 8c9845a Compare April 1, 2025 13:08

franzpoeschel added this to the 0.16.2 milestone Apr 2, 2025

franzpoeschel added bug backend: ADIOS2 labels Apr 2, 2025

franzpoeschel force-pushed the 20250328-joinedarray branch 2 times, most recently from de6838b to 9eb620f Compare April 2, 2025 13:54

franzpoeschel mentioned this pull request Apr 2, 2025

Fix double write from unique_ptr in ADIOS2 #1743

Merged

franzpoeschel force-pushed the 20250328-joinedarray branch from 9eb620f to ef62f62 Compare April 2, 2025 16:43

guj and others added 7 commits April 3, 2025 13:44

working around an unusal encounter when the joined_dim has actual

b4678bc

value "max::size_t - 1"

[pre-commit.ci] auto fixes from pre-commit.com hooks

6309c4d

for more information, see https://pre-commit.ci

Add test for redundant resetDataset()

1b3ba34

Merge check into above logic

f4dfc30

Better error messages in verifyDataset

b4af694

Add further safety guards to createDataset and extendDataset tasks

c7f896a

Move joinedDim logic into middle-end for extendDataset

240e25f

franzpoeschel force-pushed the 20250328-joinedarray branch from ef62f62 to 240e25f Compare April 3, 2025 11:44

franzpoeschel reviewed Apr 3, 2025

View reviewed changes

include/openPMD/IO/ADIOS/ADIOS2IOHandler.hpp Outdated Show resolved Hide resolved

Update include/openPMD/IO/ADIOS/ADIOS2IOHandler.hpp

154ce65

franzpoeschel enabled auto-merge (squash) April 3, 2025 11:47

franzpoeschel merged commit 0a7e0a7 into openPMD:dev Apr 3, 2025
30 checks passed

franzpoeschel mentioned this pull request Jun 3, 2025

setshape() should not touch the joined dim #1656

Closed

franzpoeschel mentioned this pull request Jun 3, 2025

Release 0.16.2: Backport non-CI PRs #1759

Merged

working around an unusual encounter when the joined_dim has actual value "max::size_t - 1" #1740

working around an unusual encounter when the joined_dim has actual value "max::size_t - 1" #1740

Uh oh!

Conversation

guj commented Mar 28, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

franzpoeschel Mar 28, 2025

Choose a reason for hiding this comment

Uh oh!

guj Mar 28, 2025

Choose a reason for hiding this comment

Uh oh!

franzpoeschel Apr 1, 2025

Choose a reason for hiding this comment

Uh oh!

guj left a comment

Choose a reason for hiding this comment

Uh oh!

guj Mar 28, 2025

Choose a reason for hiding this comment

Uh oh!

franzpoeschel commented Apr 1, 2025

Uh oh!

franzpoeschel commented Apr 1, 2025

Uh oh!

franzpoeschel commented Apr 1, 2025

Uh oh!

franzpoeschel commented Apr 2, 2025

Uh oh!

guj commented Apr 2, 2025

Uh oh!

guj commented Apr 2, 2025

Uh oh!

franzpoeschel commented Apr 2, 2025

Uh oh!

guj commented Apr 3, 2025

Uh oh!

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

guj commented Mar 28, 2025 •

edited

Loading