Gracefully handle time jumps #2255
Merged
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
eboasson
force-pushed
the
timejump-support
branch
from
f9a4f4d to
95a4b44
Compare
dpotman
reviewed
Jul 31, 2025
Contributor
dpotman
left a comment
dpotman
left a comment
There was a problem hiding this comment.
Looks good! A few minor things:
- the test code for the C++ API is using
ddsrt_cond_waitfor, so that needs an update (or a wrapper for this function in DDSRT?) sync.cfor FreeRTOS still hasddsrt_cond_waitfor
And at some places (e.g. waitset, xeventq, writer throttling) it may be helpful to add a comment with the rationale for using mtime_t or etime_t (although technically it makes no difference in many cases)
src/ddsrt/CMakeLists.txt
Outdated
| add_library(ddsrt INTERFACE) | ||
| if(ENABLE_MIMALLOC) | ||
| target_link_libraries(ddsrt INTERFACE mimalloc) | ||
| endif() |
Contributor
There was a problem hiding this comment.
Looks like a change from #2258 accidentally was included in src/ddsrt/CMakeLists.txt?
Signed-off-by: Erik Boasson <[email protected]>
Signed-off-by: Erik Boasson <[email protected]>
This is a IDL file used for testing these combinations, so disabling the warning is reasonable. Signed-off-by: Erik Boasson <[email protected]>
The only call site used an actual size larger than sizeof(char *) so this only resulted in truncation in an error path. Signed-off-by: Erik Boasson <[email protected]>
This splits out the generic ddsrt_cond_t into 4 different ones: - ddsrt_cond_t: no associated clock, can't be used with timeouts - ddsrt_cond_wctime_t: associated with ddsrt_wctime_t - ddsrt_cond_mtime_t: associated with ddsrt_mtime_t - ddsrt_cond_etime_t: associated with ddsrt_etime_t And removes the relative timeouts from the internal interface (this actually simplifies the code in several cases). For reference: - wctime is wall-clock time, a.k.a. CLOCK_REALTIME, system time, default clock, ... - mtime_t is a monotonic clock, ticking only when the machine is running (not suspended) - etime_t is a monotonic clock, ticking also when the machine is suspended Not all platforms provide all three, and perhaps on some platforms all three are present but not actually used by Cyclone DDS. Platforms like macOS where a condition variable can't be associated with a clock necessarily have to fake it by doing an approximate conversion of the timeout to the system time. Note that the generic dds_time_t has the same clock source as ddsrt_wctime_t. Signed-off-by: Erik Boasson <[email protected]>
Signed-off-by: Erik Boasson <[email protected]>
Signed-off-by: Erik Boasson <[email protected]>
Some platforms (QNX is an example) have low-resolution clocks that one can use for timeouts, and high-resolution clocks that cannot be used for that. This introduces a separate "high resolution" time that can be used only for computing time elapsed between two events. Signed-off-by: Erik Boasson <[email protected]>
Signed-off-by: Erik Boasson <[email protected]>
There is currently an assumption in the Cyclone serializers that an enum maps to an (unsigned) int and than an int is 32-bits. Because of that, printing an enum with a format string of PRIu32 will work fine, but there should be a cast to uint32_t nonetheless. Signed-off-by: Erik Boasson <[email protected]>
Signed-off-by: Erik Boasson <[email protected]>
Signed-off-by: Erik Boasson <[email protected]>
Signed-off-by: Erik Boasson <[email protected]>
eboasson
force-pushed
the
timejump-support
branch
from
95a4b44 to
3265dd8
Compare
noxpardalis
approved these changes
Aug 4, 2025
Contributor
noxpardalis
left a comment
noxpardalis
left a comment
There was a problem hiding this comment.
The new changes look good to me and address @dpotman’s points. All tests also run fine locally.
I also took a gander at the associated PR for the C++ repo: eclipse-cyclonedds/cyclonedds-cxx#546 and it looks fine. I can do a full review once this is merged and the noise from the CI failures on the C++ repo disappear on a re-run.
Contributor
Author
Thank you! |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
3 participants
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
The main commit in this PR allows blocking with timeouts based on a monotonic clock rather than the system clock. This should address the various problems with time jumps people have been facing. I haven't actually tested it with time jumps yet because of some practicalities, but perhaps some kind souls are willing to give it a go as it is.
Edit: I accidentally pushed some more onto this branch, but it fits well enough with the subject: some platforms have low-resolution clocks that interact with the scheduler but can provide a high-resolution clock that is unrelated to scheduling. Using the latter for benchmarking seems sensible.
The other commits are just housekeeping and fixing some trivialities.
Fixes #1995 ros2/rmw_cyclonedds#276