Arvados

h1. Storing and Organizing Data

Rough demo outline

# Automatic ingest from a POSIX directory to Keep

#* Ingestor's access to staging area (could be remote NFS or sshfs mount) is arranged ahead of time

#* 3rd-party's access to staging area is arranged ahead of time

#* Ingestor runs in a screen session. Command line parameters provide project (group/folder) ID and a tag that indicates "this is for *me* to ingest".

#* Someone ("3rd-party") uploads some files to the staging area via SFTP or whatever

#* 3rd-party does an API call to "ingest-notify app". This might be a short bash script culminating in a curl command. In the API call, the 3rd-party provides a label (e.g., a sample ID) and a list of files, checksums, and an arbitrary "properties" hash containing whatever the 3rd-party wants.

#* Ingest-notify app generates a "data in staging area is ready to ingest" event via API server.

#* Ingestor waits of a "data in staging area is ready to ingest" notification via API server.

#* Ingestor reads the data from the staging area and writes it into Keep (creates one collection per API call made by 3rd-party).

#* Ingestor (or arv-put on behalf of ingestor?) makes API calls while working, to indicate progress (bytes done/todo). @arvados.v1.logs.create(object_uuid=uuid_of_upload_object)@

#* In Workbench the imported Datasets appear as Collections in the designated project

#* After data has been copied into Keep, ingestor deletes the files from the staging area (if @--delete-after@ flag given).

...

# My data gets into the right project as specified by the uploader (API call)

#* How is the staging-area ↔ project mapping specified, and how/where is it encoded/stored?

...

# Subscribe to notifications (by email and/or Workbench dashboard): when files start/finish uploading; when files are shared with customer; when files are downloaded by third party

#* For now, use existing Logs table + automatic logging of create/update/delete operations + "progress" event from arv-put (see above)

#* "Show project" page shows recent activity: one progress bar for each unfinished upload, one entry for each start/finish event.

#* Dashboard page shows recent activity from all of my projects.

...

# Move/copy collections between projects (Project RX1234, or Customer X’s files), tag them in destination project with the appropriate string (e.g., sample ID) -- defaulting to existing tag used in source project (e.g., provided at time of upload).

#* UI for presenting Groups as Projects/Folders: create, view, rename, share, delete

#* UI for copying/moving objects between folders

#* How to avoid confusion about "is this one object in two places, or are there two objects?" Note GDocs has a bit of both, "My Drive" / "Shared with me" vs. regular folders

...

# Share project with other users/groups

...

# “Anyone with this secret link can view/download” mode. Enable, disable, change magic link. Use cases: browser + “wget -r”.

#* Perhaps the secret in the secret link is an ApiClientAuthorization token, belonging to the person creating the link, scoped to a single project/collection

#* How do we implement "Anonymous user, not logged in"?

...

# See log/overview of who has accessed your shared data (incl. “anonymous user” if using secret-link-to-share); when shared/unshared; when each upload started/finished -- for a single project, and across all projects

...

# Pilot alternate Workbench group/dashboard view

...

h2. Retrospective notes

* Went well - still some merge-race at the end

* Lots of branches going in

* Not a lot of merge conflicts

* Big spec change (rejecting "ingestor" story in favor of future "remote arv-put")

* Some in-sprint deployment dependency stuff (crunch+docker, websockets)

* Please tag commits with story numbers. Currently "refs #1234" works well because redmine understands it.