Introduce ros2 bag reindex #539

jhdcs · 2020-10-15T19:00:55Z

This pull request addresses issue #395.

A new verb, reindex has been added, which attempts to reconstruct a metadata.yaml file from the sqlite3 databases found within the bag.

Functionality allowing different forms of reindex for bag types has been added as well.

Readme has been updated to indicate new verb, as well as how to use it.

Unit tests for reindex have been added to rosbag2_tests.

Karsten1987

First of all, I hope you don't mind that I changed your title. That way it's a bit more obvious what this PR is doing.

Thanks for the PR, a couple of remarks though:

The reindexer has to be more generic. I found a couple of places where .db3 and sqlite3 is hardcoded. Does it make more sense to have that reindex functionality live within the sqlite3 plugin then?
As mentioned inline, while I appreciate that you follow the style, I don't think we have to introduce interfaces for the reindexer. I don't think we need that flexibility on that level.
Boost is currently kind of a show-stopper. I think we have to come up with a rcpputils solution which can iterate over files wo/ boost.

rosbag2_cpp/CMakeLists.txt

rosbag2_cpp/include/rosbag2_cpp/reindexer.hpp

rosbag2_cpp/src/rosbag2_cpp/reindexers/sequential_reindexer.cpp

Karsten1987 · 2020-10-19T20:26:05Z

rosbag2_cpp/src/rosbag2_cpp/reindexers/sequential_reindexer.cpp

+    storage_.reset();
+  }
+}
+


Suggested change

Karsten1987 · 2020-10-19T20:27:42Z

rosbag2_cpp/src/rosbag2_cpp/reindexers/sequential_reindexer.cpp

+  metadata_ = rosbag2_storage::BagMetadata{};
+
+  // This reindexer will only work on SQLite files, so this can't change
+  metadata_.storage_identifier = "sqlite3";


that has to be more generic. What happens if I try to reindex a ros1 bagfile? Or some custom storage?

Yeah, I thought sequential_reindexer was supposed to be the reindexer that worked on the sqlite files. Not sure why. I guess I'm derpy like that.

jhdcs · 2020-10-20T12:00:05Z

Made a few comments on your suggestions, thank you for those! And no problem on the name change - I'm not spectacular with titles :)

I did have one question that's been nagging me, though: Since it's theoretically possible to extract all the data that a metadata.yaml file provides, why is the file needed? Is there another format that can't store all the information needed to read it within itself?

Karsten1987 · 2020-10-20T15:47:37Z

I did have one question that's been nagging me, though: Since it's theoretically possible to extract all the data that a metadata.yaml file provides, why is the file needed?

The idea here is that the actual data payload can be huge compared to the metadata/index and distributed over multiple files. With a dedicated metadata file, we have all that data at hand without having to iterate over the payload.
For example, imaging having multiple GBs of data stored on a network drive. In order to get the metadata, you'll only have to download a tiny yaml file without ever having to open a huge file over your network.

jhdcs · 2020-10-20T18:03:41Z

Ok. Well, the good news is I've gotten the Linux/Unix reindexer working again. The bad news is I don't really have a way of testing the Windows version right now, since ROS 2 refuses to build itself on my Windows VM...

Anyhoo, now trying to figure out the best way to transplant the reindexer into the sqlite3 plugin. If I'm reading this correctly, the sqlite3 plugin is a StorageFactoryImpl, correct? I thought that was a bit lower-level than what the reindexer does... Is there another spot that I'm missing?

Otherwise, I'll try to wire up a new reindex function in the StorageFactoryImpl, and propagate the changes along. Fingers crossed...

Karsten1987 · 2020-10-20T19:01:40Z

if you want to make your life a bit easier, I guess you could directly plant that reindex logic in a python script within ros2bag. There's a python API for sqlite3 and all it needed would be a bit of logic to fail gracefully when that script is not available for other storage plugins than sqlite3.

The reason why I could see that as a feasible solution is that I don't think we need proper API for such a logic - I can't see a use-case other than just running a script to repair a broken record.

jhdcs · 2020-10-20T19:11:10Z

That... Actually might fix a LOT of problems - We wouldn't need to worry about two code paths for finding out the files within a folder in either Windows or Linux, for example. And I've already updated the get_metadata function to return QOS profiles, so as long as I can call it from Python, we're gold.

On a slightly different subject; as you've probably noticed, I realized I was using the wrong license blurb at the top of the reindex files I've added. However, the copyright lint doesn't seem to like what my employers wanted. Is there something I can do to fix this?

EDIT: Just realized you meant I can query SQLite from python directly. Even better! Should I undo the changes to the get_metadata function, then?

Karsten1987 · 2020-10-20T23:08:58Z

On a slightly different subject; as you've probably noticed, I realized I was using the wrong license blurb at the top of the reindex files I've added. However, the copyright lint doesn't seem to like what my employers wanted. Is there something I can do to fix this?

I would recommend you'd use the regular license text and change the copyright to reflect your company name. That should pass the linters - as you can see Open Robotics, Amazon and Bosch have contributions as such. For all other legal hints, I would put them in a separate doc block below the copyright header. If that's okay with your employer.

Karsten1987 · 2020-10-23T15:41:18Z

@jhdcs you might want to have a look at ros2/rcutils#249. It's still in a draft but might be the PR you're looking for the iterate through a directory.

jhdcs · 2020-10-30T18:22:46Z

Got the python version probably working, albeit unlinted. I need to write unit tests for it, which means I need to find some test databases. Should I just copy them over from the other test packages?

Also, I'm not sure how to get this to work with compression yet...

jhdcs · 2020-11-04T16:29:14Z

I've gotten the reindex operation working for uncompressed and file-compressed SQLite bags, now. Should I squash the commits? There appears to be a commit somewhere in the rebase that's confusing DCO.

jhdcs · 2020-11-04T22:05:02Z

Figured out why the test is failing - in order to support the zstd compression scheme, a python package had to be installed via pip. The particular package is called "zstandard".

If we don't want to add new python packages to the workspace, I can comment out the areas of the code that reference it. However, reindexing operations will not be able to support zstd compressed bags.

jhdcs · 2020-11-05T13:38:56Z

Reindexing for message-compressed files has been implemented. This code is ready for review.

However, we will need to figure out how we want to handle the inclusion of the zstandard package.

jhdcs · 2020-11-20T12:07:29Z

May I please have another review on this? I'm not sure why the Rpr__rosbag2__ubuntu_focal_amd64 check is failing.

jhdcs · 2020-12-01T12:40:19Z

Okay, got some time to look into it. The Rpr__rosbag2__ubuntu_focal_amd64 check is failing because the zstandard python package is not included. Which means that the file can't be opened/recompressed, and therefore the test can't find the new metadata.yaml file, and the whole thing falls down.

Do you have any recommendations for solving this? Without that zstandard package, we're locked out of working with compressed files...

emersonknapp

Hey there - sorry I have not been involved on this PR to date!

I don't think we want to install the python-based zstandard package - because it may not match up with the version being used by the C++ code. Another reason is that we may support other compression implementations, and we don't want equivalent python bindings for every one of them. The real solution around this is probably to add an endpoint to the rosbag2_py API to decompress bagfiles, having it choose the correct compression implementation under the hood to do so.

I think as a first step with this feature it would be acceptable to support reindexing uncompressed bags and print out an error message when trying to work on a compressed bag - we could have a followup ticket for Enable reindex on compressed bagfiles

If I am reading correctly, this doesn't need to actually worry about compressed messages, right? Since the type data is available in the sqlite table uncompressed.

emersonknapp · 2020-12-07T19:08:03Z

rosbag2_cpp/CMakeLists.txt

-  rosidl_typesupport_cpp
-  rosidl_typesupport_introspection_cpp
+  PUBLIC
+    ament_index_cpp


nit: don't add extra indent, removes unrelated whitespace changes

emersonknapp · 2020-12-07T19:08:08Z

rosbag2_cpp/CMakeLists.txt

 )

+


nit: remove unrelated whitespace change

emersonknapp · 2020-12-07T19:10:25Z

rosbag2_cpp/package.xml

@@ -10,7 +10,10 @@

  <buildtool_depend>ament_cmake</buildtool_depend>

+  <build_depend>libboost-filesystem-dev</build_depend>


this may be left from an old version of master? You'll probably want to revert all the changes from rosbag2_cpp, rosbag2_storage_default_plugins, rosbag2_tests, and rosbag2_transport - they don't seem related to your PR.

Additionally, I've noticed you checked in a __pycache__ directory in rosbag2_py - that should be removed as well.

emersonknapp · 2020-12-07T19:15:38Z

ros2bag/ros2bag/verb/reindex.py

+        parser.add_argument(
+            '-s', '--storage-id', default='sqlite3',
+            help="storage identifier to be used, defaults to 'sqlite3'")
+        parser.add_argument(


Both compression-format and compression-mode choices shouldn't be duplicated. Perhaps they should be factored out of record.py and this file into a common place - or you could at least pull the values out of record.py's add_argument calls into a global level variable for that module, and from ros2bag.verb.record import COMPRESSION_MODES - something like that. The duplication will inevitable get desynchronized in the future

emersonknapp · 2020-12-07T19:21:59Z

ros2bag/ros2bag/reindexer/reindex_sqlite.py

+                 compression_fmt: Literal['', 'zstd'],
+                 compression_mode: Literal['', 'none', 'file', 'message']) -> DBMetadata:
+    # Handle compression
+    if compression_fmt != '':


Suggested change

if compression_fmt != '':

if compression_fmt:

handles both None and ''. I would recommend using None rather than empty-string for optional values.

Also - this block would be much more readable as

if compression_fmt in ['none', 'message']: # open normally elif compression_fmt == 'file': # open compressed else: raise ValueError('unknown compression mode')

The "by process of elimintation !=" pattern is a lot less readable, and includes some duplicated lines.

emersonknapp · 2020-12-07T19:22:29Z

ros2bag/ros2bag/reindexer/reindex_base.py

+
+def reindex(uri: str,
+            storage_id: str,
+            compression_fmt: Literal['', 'zstd'],


I don't think this is really a Literal value - it'll require shotgun surgery if a new compression format is added. You should probably take an Optional[str] (for both fmt and mode) - the underlying logic can decide whether or not it can handle this value - overkill and brittle to handle it in the type system.

emersonknapp · 2020-12-07T19:25:19Z

ros2bag/ros2bag/reindexer/bag_metadata.py

+        return super(RosbagYamlDumper, self).increase_indent(flow, False)
+
+
+class MetadataWriter:


Considering we already have a MetadataWriter in C++, this duplicated code will be prone to rot if the metadata format changes at all - this should be a call into rosbag2_storage MetadataIO for the logic in this file

I don't think I've seen ROS code in the past calling C++ libraries from within Python code. Is there precedent for this, or is this something that will change going forward?

jhdcs · 2020-12-08T12:26:58Z

If you are concerned about duplicating functions that are within C++, should I just go back to the C++ version of the code I had originally pushed? Perhaps with a little more logic on the Python side to grab the relative file paths, and thus avoid having to use Boost?

Or should I just try to open up the C++ API to accept calls to both get the metadata, and write it out?

emersonknapp · 2020-12-08T18:54:01Z

I'm concerned with code duplication in general. I think when Karsten suggested a Python script, I don't think he intended to suggest that we rewrite existing logic in Python - just that a Python top level script would be easier to manage than fully-C++ logic.

We definitely do not want to introduce a boost dependency, we intentionally removed it from the ros2 core and rosbag2 in most places. For path manipulation, the C++ stdlib handles a lot of that now, and there are a good number of shims for any missing functions in https://github.com/ros2/rcpputils

I personally think that expanding the rosbag2_py API to expose metadata operations would be a desirable feature in and of itself, and could be a standalone PR that would enable this feature

danthony06

I think that overall the changes look good, and it would be nice to get this functionality merged into future releases.

jhdcs · 2021-01-04T14:47:50Z

I'm concerned with code duplication in general. I think when Karsten suggested a Python script, I don't think he intended to suggest that we rewrite existing logic in Python - just that a Python top level script would be easier to manage than fully-C++ logic.

We definitely do not want to introduce a boost dependency, we intentionally removed it from the ros2 core and rosbag2 in most places. For path manipulation, the C++ stdlib handles a lot of that now, and there are a good number of shims for any missing functions in https://github.com/ros2/rcpputils

I personally think that expanding the rosbag2_py API to expose metadata operations would be a desirable feature in and of itself, and could be a standalone PR that would enable this feature

I could revert all of the python changes and go back to the C++ version. However, that still leaves us with the problem that I can't query what files are in a directory. rccputils doesn't have that functionality in its filesystem_helper.hpp file, and std::filesystem doesn't exist in C++14. It does in C++17, but I'm not sure if or when ros2 will be switching to that version.

The only other possibility I can think of that doesn't add a new dependency is to figure out how to have Python send a list of strings to C++, and have it converted into a std::vector. That way Python can query what files are in the directory, and give it to C++ for further processing. But I have no idea how to do that...

emersonknapp · 2021-01-14T21:24:01Z

Yes - I am currently in the process of refactoring the compression code pretty seriously. Like I said in an earlier comment on this review, though, I think it's perfectly acceptable to partially implement this feature (not supporting compressed bags for a first pass).

As a note - I am finding this PR difficult to review due to the sheer size of it - and would prefer smaller PRs that implement individual atomic functionalities.

For example - if you are introducing a SequentialReindexer class that can do the reindexing functionality - then a PR introducing just that class, with tests and a usable C++ API would be a great PR that I'll be much more likely to be able to review effectively. I'm looking for PRs that are self-contained, demonstrably correct, and do one usable thing well, even if it doesn't do everything you would want it to do. That's something that we could really make meaningful progress on.

ros-discourse · 2021-01-18T12:21:30Z

This pull request has been mentioned on ROS Discourse. There might be relevant details there:

https://discourse.ros.org/t/tooling-wg-breakout-meeting-2021-01-22-rosbag2-reindexing/18489/3

Signed-off-by: Jacob Hassold <[email protected]>

Distro A, OPSEC #4584 Signed-off-by: Jacob Hassold <[email protected]>

Karsten1987

First off, thanks a lot for keeping this PR alive, I know it's not trivial. A few things here:

I would try to keep the change to a minimal set. That is, if possible, I would prefer if we'd started without compression and end-to-end integration to fix the internal architecture of reindexing. Similarly, I was wondering what's the purpose of introducing an interface for the indexer. Don't get me wrong here, there might be a legit reason for it, I just want to rule out that we don't introduce unnecessary complexity. That is, I believe a reindexing of the database will always yield the same result, I don't really see what's a sequential reindexing doing differently from a different method of reindexing.
We've currently introduced a set of functions within rcutils (linked inline) which replaces much of the custom logic to iterate through files in a directory. I strongly advertise this.
I feel there's a lot of code around file handling, rosbag versioning and such which could be reduced or reorganized to be reused across the rosbag2 functionality. I believe there's code within the sequential_indexer wich can be reused within the sequential reader and writer.
I don't really like that rosbag2_transport is involved in this. I know it's currently our only entry-point from ros2bag to rosbag2_cpp, but would it make sense to rather go through rosbag2_py instead? I don't think it's a hard requirement here for me, but I think we should generally rethink the architecture on how to deal with python and cpp, now that we have rosbag2_py

Karsten1987 · 2021-01-22T16:15:20Z

rosbag2_cpp/src/rosbag2_cpp/reindexers/sequential_reindexer.cpp

+  #ifdef WIN32
+  {
+    // Code placed in scope so variables can't accidentally be used elsewhere
+    WIN32_FIND_DATA FindFileData;
+    HANDLE hFind;
+
+    std::string search_dir = base_folder.string() + "\\*";
+    hFind = FindFirstFile(search_dir, &FindFileData);
+    // If an error occurs, we want to abort
+    if (hFind == INVALID_HANDLE_VALUE) {
+      DWORD dwError = GetLastError();
+      std::error_code ec(dwError, std::system_category());
+      throw(std::system_error(ec));
+    }
+
+    // Loop through the directory, collecting file names as we go
+    do {
+      if (ffd.dwFileAttributes & FILE_ATTRIBUTE_DIRECTORY) {
+        // I guess it's a directory in the bag file?
+        // Still, not interested in it.
+        continue;
+      } else {
+        auto temp_path = rcpputils::fs::path(ffd.cFileName);
+
+        // We are ONLY interested in database files
+        if (temp_path.extension().string() != ".db3") {
+          continue;
+        }
+        output.emplace_back(temp_path);
+      }
+    } while (FindNextFile(hFind, &ffd) != 0);
+
+    FindClose(hFind);
+  }
+  #else
+  {
+    // Code placed in scope so variables can't accidentally be used elsewhere
+    auto dirp = opendir(base_folder.string().c_str());
+    // If an error occurs, we want to abort
+    if (dirp == NULL) {
+      throw std::system_error(errno, std::generic_category());
+    }
+    dirent * dp;
+    while ((dp = readdir(dirp)) != NULL) {
+      auto non_const_folder = rcpputils::fs::path(base_folder);
+      auto temp_path = non_const_folder /= rcpputils::fs::path(dp->d_name);
+
+      // We are ONLY interested in database files
+      if (temp_path.extension().string() != ".db3") {
+        continue;
+      }
+
+      output.emplace_back(temp_path);
+    }
+    closedir(dirp);
+  }
+  #endif


We've recently added code which supersedes this: ros2/rcutils#323

Karsten1987 · 2021-01-22T16:15:42Z

rosbag2_cpp/src/rosbag2_cpp/reindexers/sequential_reindexer.cpp

+// We're on a UNIX system. Import their filesystem stuff instead
+#include <dirent.h>
+#endif
+


Suggested change

Karsten1987 · 2021-01-22T16:15:51Z

rosbag2_cpp/src/rosbag2_cpp/reindexer.cpp

+{
+  reindexer_impl->reindex(storage_options);
+}
+


Suggested change

Karsten1987 · 2021-01-22T16:16:05Z

rosbag2_cpp/package.xml

@@ -11,6 +11,7 @@
  <buildtool_depend>ament_cmake</buildtool_depend>

  <depend>ament_index_cpp</depend>
+  <depend>libboost-filesystem</depend>


Suggested change

<depend>libboost-filesystem</depend>

Karsten1987 · 2021-01-22T16:16:22Z

rosbag2_cpp/include/rosbag2_cpp/reindexers/sequential_reindexer.hpp

+    const std::vector<rcpputils::fs::path> & files,
+    const rosbag2_storage::StorageOptions & storage_options);
+
+  // Compairson function for std::sort with our filepath convention


Suggested change

// Compairson function for std::sort with our filepath convention

// Comparison function for std::sort with our filepath convention

Karsten1987 · 2021-01-22T16:16:42Z

rosbag2_cpp/include/rosbag2_cpp/reindexers/sequential_reindexer.hpp

+    std::make_unique<rosbag2_storage::MetadataIo>());
+
+  virtual ~SequentialReindexer();
+


Suggested change

Karsten1987 · 2021-01-22T16:25:34Z

rosbag2_cpp/src/rosbag2_cpp/reindexers/sequential_reindexer.cpp

+{
+namespace reindexers
+{
+namespace details


I feel like the following two functions should be accessible independently from the reindexer. Aren't we having similar code in the reader already? I think we're introducing code duplication here.
Especially, given that version differentiation, I believe it's important to keep code as such to a minimum.

Is there a spot that you think this would fit better in?

Also, looking back on the code, not sure why I have the version parameter there. Huh. That should probably come out.

Karsten1987 · 2021-01-22T16:29:17Z

rosbag2_storage_default_plugins/src/rosbag2_storage_default_plugins/sqlite/sqlite_storage.cpp

@@ -430,19 +430,19 @@ rosbag2_storage::BagMetadata SqliteStorage::get_metadata()

  auto statement = database_->prepare_statement(
    "SELECT name, type, serialization_format, COUNT(messages.id), MIN(messages.timestamp), "
-    "MAX(messages.timestamp) "
+    "MAX(messages.timestamp), offered_qos_profiles "


why are we needing this change here for the reindexing?

This is because the original metadata file includes the QOS profiles of the topics, so I have to query them from the database. I did some searching around, and I don't think this change affects anything else in the project, though if you have an alternate way to query the QOS profiles from the database, that might be preferable.

jhdcs · 2021-02-08T15:23:31Z

Closing in favor of #641

jhdcs requested a review from a team as a code owner October 15, 2020 19:00

jhdcs requested review from emersonknapp and Karsten1987 and removed request for a team October 15, 2020 19:00

jhdcs force-pushed the bug/missing_metadata branch from f568738 to eaabdc0 Compare October 15, 2020 19:27

Karsten1987 changed the title ~~Bug/missing metadata~~ Introduce ros2 bag reindex Oct 19, 2020

Karsten1987 reviewed Oct 19, 2020

View reviewed changes

jhdcs marked this pull request as draft October 20, 2020 14:47

jhdcs marked this pull request as ready for review November 4, 2020 16:26

jhdcs force-pushed the bug/missing_metadata branch from d7cb035 to 99509f8 Compare November 4, 2020 20:47

jhdcs requested a review from Karsten1987 November 18, 2020 19:24

emersonknapp requested changes Dec 7, 2020

View reviewed changes

danthony06 reviewed Dec 14, 2020

View reviewed changes

jhdcs force-pushed the bug/missing_metadata branch from 25a6814 to d499aeb Compare January 15, 2021 12:27

Jacob Hassold added 18 commits January 18, 2021 08:54

Distro A, OPSEC #4584

55a3d03

Signed-off-by: Jacob Hassold <[email protected]>

Remove Boost

3d3afba

Distro A, OPSEC #4584 Signed-off-by: Jacob Hassold <[email protected]>

Remove extra newline

b26db0a

Distro A, OPSEC #4584 Signed-off-by: Jacob Hassold <[email protected]>

Change exception to be in line with Unix

3153d81

Distro A, OPSEC #4584 Signed-off-by: Jacob Hassold <[email protected]>

Fixed syntax error

1654cd5

Distro A, OPSEC #4584 Signed-off-by: Jacob Hassold <[email protected]>

Linux reindexer now working

28c48a4

Distro A, OPSEC #4584 Signed-off-by: Jacob Hassold <[email protected]>

FIXED USING WRONG LICENSE

94acad2

Distro A, OPSEC #4584 Signed-off-by: Jacob Hassold <[email protected]>

Remove sneaky boost reference

5e533ff

Distro A, OPSEC #4584 Signed-off-by: Jacob Hassold <[email protected]>

Test reordering license header

0444bd4

Distro A, OPSEC #4584 Signed-off-by: Jacob Hassold <[email protected]>

Test another license reordering

d161c03

Distro A, OPSEC #4584 Signed-off-by: Jacob Hassold <[email protected]>

Convert to rosbag2_storage::StorageOptions

eea8638

Distro A, OPSEC #4584 Signed-off-by: Jacob Hassold <[email protected]>

Missed a StorageOptions in the previous commit

b5e7f10

Distro A, OPSEC #4584 Signed-off-by: Jacob Hassold <[email protected]>

Add test data

9b6d8ec

Distro A, OPSEC #4584 Signed-off-by: Jacob Hassold <[email protected]>

Ignore __pycache__ in commits

5152360

Distro A, OPSEC #4584 Signed-off-by: Jacob Hassold <[email protected]>

Re-use Python tests

1495ddd

Distro A, OPSEC #4584 Signed-off-by: Jacob Hassold <[email protected]>

Implement Unit Testing

bb0d5af

Distro A, OPSEC #4584 Signed-off-by: Jacob Hassold <[email protected]>

Fixes that were suggested but buried in PR

caeb348

Distro A, OPSEC #4584 Signed-off-by: Jacob Hassold <[email protected]>

Fix init_metadata function call

a2d6321

Distro A, OPSEC #4584 Signed-off-by: Jacob Hassold <[email protected]>

jhdcs force-pushed the bug/missing_metadata branch from 5e5798a to a2d6321 Compare January 18, 2021 13:55

Jacob Hassold added 3 commits January 18, 2021 09:11

Add missing reindexer_ parameters

b5f74f8

Distro A, OPSEC #4584 Signed-off-by: Jacob Hassold <[email protected]>

Disable compression testing

7d43d3f

Distro A, OPSEC #4584 Signed-off-by: Jacob Hassold <[email protected]>

Linting fixes

f47aba9

Distro A, OPSEC #4584 Signed-off-by: Jacob Hassold <[email protected]>

Karsten1987 reviewed Jan 22, 2021

View reviewed changes

jhdcs mentioned this pull request Feb 8, 2021

Introducing Reindexer core #641

Merged

jhdcs closed this Feb 8, 2021

		@@ -10,7 +10,10 @@

		<buildtool_depend>ament_cmake</buildtool_depend>

		<build_depend>libboost-filesystem-dev</build_depend>

		return super(RosbagYamlDumper, self).increase_indent(flow, False)


		class MetadataWriter:

	// Compairson function for std::sort with our filepath convention
	// Comparison function for std::sort with our filepath convention

		std::make_unique<rosbag2_storage::MetadataIo>());

		virtual ~SequentialReindexer();

Introduce ros2 bag reindex #539

Introduce ros2 bag reindex #539

Conversation

jhdcs commented Oct 15, 2020

Karsten1987 left a comment • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

jhdcs commented Oct 20, 2020 • edited Loading

Karsten1987 commented Oct 20, 2020 • edited Loading

jhdcs commented Oct 20, 2020

Karsten1987 commented Oct 20, 2020 • edited Loading

jhdcs commented Oct 20, 2020 • edited Loading

Karsten1987 commented Oct 20, 2020

Karsten1987 commented Oct 23, 2020

jhdcs commented Oct 30, 2020

jhdcs commented Nov 4, 2020

jhdcs commented Nov 4, 2020

jhdcs commented Nov 5, 2020

jhdcs commented Nov 20, 2020

jhdcs commented Dec 1, 2020

emersonknapp left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

jhdcs commented Dec 8, 2020 • edited Loading

emersonknapp commented Dec 8, 2020

danthony06 left a comment

Choose a reason for hiding this comment

jhdcs commented Jan 4, 2021

emersonknapp commented Jan 14, 2021

ros-discourse commented Jan 18, 2021

Karsten1987 left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

jhdcs commented Feb 8, 2021

Karsten1987 left a comment •

edited

Loading

jhdcs commented Oct 20, 2020 •

edited

Loading

Karsten1987 commented Oct 20, 2020 •

edited

Loading

Karsten1987 commented Oct 20, 2020 •

edited

Loading

jhdcs commented Oct 20, 2020 •

edited

Loading

jhdcs commented Dec 8, 2020 •

edited

Loading