marius a. eriksen

PYLS has two quirks: first, it requires that window/showStatus is implemented; second, on formatting it replaces the full text, which causes a bad user experience in acme. This PR addresses...

Support direct access to s3, etc., through URL schemes.

1

Applications that support directly accessing S3 or other cloud storage providers could be sped up (sometimes by a lot) by avoiding staging which is currently necessary in order to make...

Reflowlet currently uses port 9000 for reflow->reflowlet and inter-reflowlet communications. This can be problematic because many firewalls block unusual ports. Instead, we should use the standard SSL port (443).

Currently, if a bucket's region is undiscoverable (e.g., because we don't have permissions to issue `GetBucketLocation` calls), Reflow reverts to the default region `us-east-1`. Reflow should probably instead use the...

Reflow provides various ways to inspect running and completed tasks, e.g., reflow info, reflow logs, reflow ps, etc.; these should work while operating in local mode as well.

Docker image resolver is too strict

4

Trying to run a simple Reflow program with the image "biocontainers/bwa" fails: ``` $ reflow run align.rf obtaining manifest digest for "index.docker.io/biocontainers/bwa:latest": MANIFEST_UNKNOWN: "manifest unknown" $ ```

Reflow should support slicing lists (e.g., `l[:100]`) as in Go.

Reflow should automatically restart jobs that encounter zombie allocs. See: https://github.com/grailbio/reflow/issues/43

Currently, when a continuation node is forked, it is not canonicalized. This is done mostly due to simplicity, since doing this properly would require us to rewrite pointers _to_ the...

"keep going" mode

1

Users should be given the option to try to complete as much of a Reflow program in the case of a failure. This is useful for programs that compute many...