What's the best way to refresh feeds faster with the Feeds module?

Yes, thanks!

Do I need to write a new FeedsScheduler that will add the feed to the queue table faster than cron() could?

Correct me if I'm wrong, here's how I see Feeds + Drupal Queue works:
* cron() adds feeds_queue to Drupal Queue
* Drupal Queue will remove feeds_queue once it is "drush queue cron" is ran.

If I want to refresh a particular Feed Importer every 10 seconds, how do I add it to Drupal Queue faster than cron() hook could? Do I need to, is there another way?

I created some drush commands for the feeds module. (The file is coded for Drush 3.)

The commands available are:

drush feeds-config
* Displays all active importers or displays the config of a given importer (passed as arg).

drush feeds-refresh
* Refreshes a feed based on its schedule.

drush feeds-queue
* Adds a scheduled feed to the drupal_queue. (Needs to be run in conjunction with "drush queue cron".)

I will work on the shell script that will execute these drush commands.

I'm fetching NewsML (an xml standard for multimedia news) files from a server via SSH. If a newsml file popups up in the server, the feeds module has to fetch, parse, and process it RIGHT AWAY in the drupal site (in less than 60 seconds).

For this, I had to create these plugins:
* SSHFetcher
* NewsMLParser
* NewsMLProcessor

As a side note, the NewsML has a very different structure than RSS/syndication xmls. That's why the FetcherResult object in Alpha9 suited me really well; it has no specified structure. With FeedsImportBatch, the only variable I'm using is the $items. Its getRaw() does not quite work for me because I'm fetching it via SSH not http.

The logic was inside fetch()... creating an SSHBatchImport is much better, and I can still ensure that all SSH connections only happen in the fetch phase.

Thanks for the tip!!

My use case calls for high availability, and the news feeds need to be parsed instantly. I had to create a feeds-refresh command outside of the drupal_queue to avoid cases where there's a lot of other jobs in the queue, which might slow down the news feed parsing for this particular importer.

If the queue has a mechanism that will push my high-priority job to the top, then I would use that. Right now, it's easier to create a dedicated process to handle the importing.

Additional question: when "drush queue cron" is run, does it fire-off the callbacks sequentially or in parallel? How does it determine which to run first?

It's not high volume. On average, about 20 items every 10 seconds. Plus, I'm capping the fetcher to only get 100 items per cycle.

What's the benchmark on the Feeds module on how much feed items it can fetch, parse, process without timing out or breaking down? With my testing, the feeds module was able to process 250 feed items (24kb each) in under 40 seconds. So I'm assuming 100 items per cycle is still safe.

Thanks for the drush queue explanation. So it is possible to run parallel executions.