I am curious what the max number of feeds on average people are able to get without killing drupal or having chron runs time out. Whats the best strategy when you have a hundred or so feeds?

Comments

alex_b’s picture

That's a very difficult question to answer globally.

Overall, aggregation performance depends on:

- Your server's CPU and storage I/O performance
- Your server's network connection
- The content type being created (complex CCK content type? simple Data record?).
- The activity of your feeds being processed (many new items per run?).
- The number of feeds being processed.
- The parser being used (not as critical as other factors).

Usually, as performance degrades you will see feeds that appear to be stale (no new items present although original feed has been updated a while ago). The staleness will increase with the number of feeds you add. A good measure of overall aggregation performance is the time difference between the most recently updated feed and the last updated feed:

# my_importer_id is the id of the importer to be examined (can be looked up in feeds_importer table).
SELECT MAX(last_executed_time) - MIN(last_executed_time) FROM feeds_schedule WHERE id = 'my_importer_id';
alex_b’s picture

Status: Active » Fixed

I have added #1 to the docs: http://drupal.org/node/622698

Can we consider your question answered?

AntiNSA’s picture

Sure.. I was hoping people would tell me their averages... I am getting php time outs with like 15-20 feed sources.... I hope other people will leave their feed source amounts so I can get a fair idea....

I am looking into yahoo pipes as a way to offload the stress on the server.

alex_b’s picture

Status: Fixed » Active

Hang on, you shouldn't get time outs with 15-20 feeds. You shouldn't get timeouts at all.

- What else are you running on cron (search?)
- Any chance to share the feeds here?
- Are you running on a shared server?
- Have you tried using drush to run cron?
- How do the timeouts manifest themselves? How do you notice that there are timeouts?

bflora’s picture

Also wondering about performance.

I'm running Feeds on a 4GB server and often get timeouts during the initial import of nodes when I create a new feed.

I'm importing 10-12 youtube feeds. Am using pubsubhub and telling it to wait at least 30 mins before updating again. When I import a node, I'm filling in the embedded media field to create a "video" content type. I'm also shipping the nodes out to Open Calais for automated tagging.

The setup works, but I've noticed my server crapping out on me 10-12 times a day since I started doing this. Not good, obviously.

Any tips on ways to identify where the problem might lie?

Would I improve performance by telling pubsubhub to wait longer before importing stuff?

This is a great module, but like all Drupal modules I've seen that rely on Drupal to import stuff, the "black box" nature of the system makes it hard to ever trust it.

alex_b’s picture

Title: What are the maxinum number of feeds you are able to have with feeds on average without bogging down drupal or causing timeouts » Performance - max number of feeds

Simplifying title for better legibility of list view of issue queue.

alex_b’s picture

If all of your feeds support pubsubhubbub you can lower your refresh period from 30 minutes to somewhere like 2 days or 7 days.

Yes, PuSH feeds should be slowed down automatically, but we're waiting on #721428: Make scheduler next scheduled time based to make this happen.

Also: could you explain better what "crapping out" means in this context?

bflora’s picture

Crapping out = Apache dies and needs to be restarted.

In truth, we're still looking to pinpoint the problem, so it can't necessarily be laid at Feeds feet. That's why I'm trying to get a sense of how resource intensive the module is.

alex_b’s picture

I see - 10-12 feed subscriptions doesn't sound bad at all.

alex_b’s picture

Status: Active » Fixed

I added this question to the ongoing FAQ collection #836618: Compile FAQs

Thanks for your input guys.

Status: Fixed » Closed (fixed)

Automatically closed -- issue fixed for 2 weeks with no activity.