Thoughts on enabling support for s3/remote storage?

[pbrumm]

I recently implemented wal-e for postgres, and it's s3 storage and restore are very compelling.

[basak]

I have no problem in principle. An issue might be that although individual chunks work well with S3's object store paradigm, the sqlite3 database that maintains chunk sequencing does not. An initial implementation might have to first download the sqlite3 database to a temporary directory, work on it there, upload chunks from S3 such that the S3 archive maintains integrity, upload the modified sqlite3 database back up to S3, and then delete any chunks at that point (to maintain integrity).

Further changes might be to support per-chunk HMAC and encryption.

But I don't think this is trivial. No objection in principle, though.