Migration

Listed here are all versions that necessitate migration. Depending on the version you are migrating from you might need to follow multiple migration steps.

In case of database migrations it is sufficient to marv dump the database with the version you are currently using and marv restore with the latest version; marv is able to dump itself and restore any older version. In case this does not hold true marv restore will complain and provide instructions what to do.

Upcoming (unreleased)

21.10.0

Previously MARV shipped a fork of the upstream rosbag Python library for ROS1 bag reading and writing. These tasks are now handled by rosbags which offers better performance and a unified interface for ROS1 and ROS2 bags. One of the main user facing changes is that ROS1 message type names are now normalized to their ROS2 counterparts. While searching a database with ROS1 and ROS2 bags for images required filtering for message type sensor_msgs/Image and sensor_msgs/msg/Image in the past, the same query is now achieved just by filtering for sensor_msgs/msg/Image.

Bagmeta node for ROS1 bags needs to be rerun

Rerun the node marv_robotics.bag:bagmeta to get normalized message type names:

marv run --col="*" --force \
  --node bagmeta \
  --node summary_keyval \
  --node bagmeta_table \
  --node connections_section

Stop using classic ROS1 message type names

Update any code or API calls using classic ROS1 message type names to the corresponding ROS2 names.

Update any code using get_message_type()

MARV nodes now have access to rosbag version agnostic types, you need to update your type deserialization code.

Replace the pattern

from marv_robotics.bag import get_message_type
...
rosmsg = get_message_type(stream)() if stream.msg_type else None
...
rosmsg.deserialize(msg.data)

with

from marv_robotics.bag import make_deserialize
...
deserialize = make_deserialize(stream)
...
rosmsg = deserialize(msg.data)

21.07.0

Switch to rosbags library

Previously MARV shipped with a builtin ROS2 bag reader. Out of the box it supported deserialization of all default ROS2, Autoware.Auto, and automotive_autonomy messages. The ROS2 reading code was spun off into rosbags. To allow for more flexibility in regards to type versioning, rosbags includes a dynamic type system that can be extended during runtime and does only include ROS2 default messages.

If you are using the previously included messages in your ROS2 bags, you have to provide their message definitions to MARV with the following commands in your site folder:

mkdir -p resources/messages
cd resources/messages
git clone https://gitlab.com/autowarefoundation/autoware.auto/autoware_auto_msgs.git
git clone https://github.com/astuff/automotive_autonomy_msgs.git

Connections_section migration (EE)

The EE node marv_ee_nodes.detail:connections_section has gained support for the download of time slices. On installations that use this node its output needs to be regenerated:

marv run --force --node connections_section --col=\*

21.05.0

Code migration

• Removed deprecated marv.types, use marv_api.types instead

• Removed deprecated marv.utils.popen, use marv_api.utils.popen instead

• Removed deprecated HTTP listing API, query Collection instead

• Removed deprecated list config function, use makelist instead

Configuration migration (CE)

The acl configuration key for setting a route-based permission profile was removed. If you were previously using the marv_webapi.acls:public profile in your CE installation, you can grant anonymous users readonly access with the ce_anonymous_readonly_access (CE) configuration key.

Database migration

Dataset and leaf permission management required changes to database schemas, a migration of the MARV database is necessary. Export the database with your current version of MARV:

marv dump dump-2103.json
mv db/db.sqlite db/db.sqlite.2103

After updating MARV run:

marv init
marv restore dump-2103.json

21.03.0

Database migration

Extended user and leaf management, among others, required changes to database schemas, a migration of the MARV database is necessary. Export the database with your current version of MARV:

marv dump dump-2012.json
mv db/db.sqlite db/db.sqlite.2012

After updating MARV run:

marv init
marv restore dump-2012.json

Leaves storage configuration changes (EE)

MARV now uses a dedicated directory to store datasets uploaded by leaves (default ./leaves in your site). Update your reverse proxy configuration to serve these files. For nginx add:

location /docker/container/path/to/leavesdir {
  internal;
  alias /host/path/to/leavesdir;
}

See also Gunicorn behind NGINX.

20.12.0

Database migration

Extended RPC filters, among others, required changes to database schemas, a migration of the MARV database is necessary. Export the database with your current version of MARV:

marv dump dump-2008.json
mv db/db.sqlite db/db.sqlite.2008

After updating MARV run:

marv init
marv restore dump-2008.json

Code migration (CE)

Node marv_robotics.trajectory.navsatfix() now returns timestamps as integer in nanoseconds instead of as float in seconds. Directly consuming custom nodes need migration, all nodes shipping with marv are already adjusted accordingly.

Config migration (EE)

To implement partial downloads, the Enterprise Edition now uses a dedicated connections section. Please adjust your config accordingly and rerun the corresponding node:

-    marv_robotics.detail:connections_section
+    marv_ee_nodes.detail:connections_section

marv run --force --node connections_section --col=\*

20.08.0

Session key file

A marv site contains a session key file used in authenticating marv users. This file was previously readable by all users of the host system. We recommend to delete the file and restart marv to recreate it automatically. All marv users will have to relogin.

rm site/sessionkey

Table column sorting

Table columns containing links use dictionaries instead of plain values and the sort order of these will seem random. Use sortkey on table columns to define the dictionary key for sorting (see Table for more information).

In case you are using marv_robotics.detail.bagmeta_table() and have datasets with multiple files, you might want to run:

marv run --node bagmeta_table --force --col=*

Likewise for marv_nodes.meta_table().

20.06.0

Changes to access control

MARV supports access control profiles that dictate who is permitted to perform what. Profiles map actions like comment or tag to lists of user groups that are permitted to perform these actions. This version of MARV cleans up and streamlines the list of supported verbs.

If you created a custom profile, please refer to marv_webapi.acls as an example and make sure to adjust your custom profile to use the new action verbs.

No migration is necessary for installations that use one of the two default profiles (authenticated or public).

Database migration

Changes to collection management required changes to database schemas. A migration of the MARV database is necessary. Export the database with your current version of MARV:

marv dump dump-2004.json
mv db/db.sqlite db/db.sqlite.2004

After updating MARV run:

marv init
marv restore dump-2004.json

20.04.0

Database migration

An updated version of tortoise-orm required changes to the database schemas. A migration of the MARV database is necessary. Export the database with your current version of MARV:

marv dump dump-1911.json
mv db/db.sqlite db/db.sqlite.1911

After updating MARV run:

marv init
marv restore dump-1911.json

19.11.0

The gunicorn configuration was simplified. Instead of providing gunicorn_cfg.py and running gunicorn manually, the marv serve cli was added. Check it out with marv serve --help.

If you were overriding the dburi path in your marv.conf, there is no need for the odd sqlalchemy URIs with four slashes anymore. If your custom dburi starts with sqlite://// please remove one slash.

This version makes the switch from sqlalchemy to tortoise as the underlying ORM, which makes a migration of the MARV database necessary. Export the database with your current version of MARV:

marv dump dump-1909.json
mv db/db.sqlite db/db.sqlite.1909

After updating MARV run:

marv init
marv restore dump-1909.json

19.09.0

Uwsgi was replaced in favour of Gunicorn. In your site directory, replace uwsgi.conf with a gunicorn_cfg.py:

"""Gunicorn configuration for MARV."""

import multiprocessing
import pathlib

# pylint: disable=invalid-name

bind = ':8000'
proc_name = 'marvweb'
raw_env = {
    f'MARV_CONFIG={pathlib.Path(__file__).parent.resolve() / "marv.conf"}',
}
workers = multiprocessing.cpu_count() * 2 + 1
worker_class = 'aiohttp.GunicornUVLoopWebWorker'

If you made any changes to your old uwsgi.conf please adjust the above config accordingly. If you are using an nginx reverse proxy you also have to adjust its configuration. Replace any uwsgi_pass directives with a proxy_pass:

-       uwsgi_pass 127.0.0.1:8000;
-       include uwsgi_params;
+       proxy_set_header Host $host;
+       proxy_set_header X-Forwarded-For $proxy_add_x_forwarded_for;
+       proxy_pass 127.0.0.1:8000;

19.02.0

Support for ipdb has been dropped in favour of pdb++. Use PDB=1 marv run instead of marv-ipdb run. For more information see Debugging.

18.07

The way inputs are handled has changed. Inputs selecting an individual topic are now optional. See Optional inputs for more information.

18.04

The list of system dependencies is updated and the installation has significantly changed. We recommend that you re-read the Installation instructions. The database has not changed and existing sites continue to function without migration.

MARV now supports offloading the delivery of files to nginx. In case you are not using nginx as reverse-proxy, yet, you should seriously consider to change that now. See reverse_proxy and Gunicorn behind NGINX.

MARV now supports access control lists (ACLs). The default ACL requires authentication to read anything, tag and comment; and only members of the group admin may discard datasets. For users of the Enterprise Edition this corresponds to the same behaviour as before.

18.03

With this release:

• geojson property object has changed.

To update the store to the new format rerun the trajectory nodes using:

marv run --node trajectory --force --force-dependent --collection=*

18.02

With this release:

• message definitions are read from bag files instead of being expected on the system

• marv allows mixing message types per topic.

As part of this the topics section has been renamed to connections_section and the bagmeta node has changed a bit. Please update your configuration along the lines of:

diff --git a/docs/config/marv.conf b/docs/config/marv.conf
index 0e44b19..31dfe57 100644
--- a/docs/config/marv.conf
+++ b/docs/config/marv.conf
@@ -24,7 +24,7 @@ nodes =
     marv_robotics.detail:summary_keyval
     marv_robotics.detail:bagmeta_table
     # detail sections
-    marv_robotics.detail:topics_section
+    marv_robotics.detail:connections_section
     marv_robotics.detail:images_section
     marv_robotics.detail:video_section
     marv_robotics.detail:gnss_section
@@ -43,8 +43,8 @@ filters =
     start_time | Start time    | lt le eq ne ge gt | datetime   | (get "bagmeta.start_time")
     end_time   | End time      | lt le eq ne ge gt | datetime   | (get "bagmeta.end_time")
     duration   | Duration      | lt le eq ne ge gt | timedelta  | (get "bagmeta.duration")
-    topics     | Topics        | any all           | subset     | (get "bagmeta.topics[:].name")
-    msg_types  | Message types | any all           | subset     | (get "bagmeta.msg_types[:].name")
+    topics     | Topics        | any all           | subset     | (get "bagmeta.topics")
+    msg_types  | Message types | any all           | subset     | (get "bagmeta.msg_types")
 
 listing_columns =
     # id       | Heading    | formatter | value function
@@ -69,7 +69,7 @@ detail_summary_widgets =
     bagmeta_table
 
 detail_sections =
-    topics_section
+    connections_section
     images_section
     video_section
     gnss_section

And then rerun the bagmeta node and the new connections section.

marv run --node bagmeta --node connections_section --force --collection=*

Now, nodes can be run, that were previously missing message type definitions. gnss_plots for example works differently, if it cannot find navsat orientations. To rerun it and all nodes depending on it:

marv run --node gnss_plots --force --force-dependent --collection=*

17.11

In the old site:

1. Make sure you have a backup of your old site!

2. Dump database of old marv:

   curl -LO https://gist.githubusercontent.com/chaoflow/02a1be706cf4948a9f4d7f1fd66d6c73/raw/de4feab88bcfa756abfb6c7f5a8ccaef7f25b36d/marv-16.10-dump.py
   python2 marv-16.10-dump.py > /tmp/dump.json
   

For and in the new instance:

3. Follow Installation and Tutorial: Setup basic site to setup a basic site.

4. Replace marv.conf with the default Configuration and adjust as needed (e.g. scanroot).

5. Initialize site with new configuration:

   marv init
   

6. If your scanroot has moved, adjust paths as needed:

   sed -i -e 's,/old/scanroot/,/path/to/new/scanroot/,g' /tmp/dump.json
   

7. Restore database in new marv:

   marv restore /tmp/dump.json
   

8. Set password for each user:

   marv user pw <username>
   

9. Run nodes:

   marv query -0 --collection=bags |xargs -0 -L25 -P4 marv run --keep-going
   

10. Run again sequentially to see if there are nodes producing errors:

marv run --collection=bags