And again, Sahara when it comes to new MCM tasks. Only one resend in over 2 hours.

There are a couple of replies to your equivalent query in the ARP1 "New work" thread -- I'd posted there before I saw this -- Tony Ellis and I can account for a whole 9 new WUs between us :-)
I'd been wondering about that, because until recently the new work droughts seemed to be less frequent...

I'll try to make time to have a closer look at my returned results timelines over the last few weeks to look at tasks with retries (either processed or awaited) -- the old pattern was that after a certain amount of work had gone out, it seemed that missed deadline tasks would start to back up (suggesting that tasks were possibly going out in descending WU order?) until action was taken, but more recently there don't seem to be many retries held up but a [complete] lack of new work while they're being dealt with has become a more frequent occurrence.

Unless there are huge numbers of retries waiting, simply sifting in ascending WU order wouldn't explain the lack of new work, whereas turning off some part of the path between creating the data files for a WU and letting the database provide it to the feeder would, so (as per your comment) I have also tended to assume that they'd been turning something off (which only seemed to push the problem out for another cyclic repeat!)

It will be interesting to see if they can solve this before MAM1 goes live -- as I seem to recall some comments about them having problems quite some time ago when an uncommon host requirement blocked the feeders I don't know that they'll be able to use a simple "ascending WU order" method -- there's supposed to be an option that avoids server cache blockages by using slots for different HR classes, but that had probably been tried already so it presumably wasn't helping!

Cheers - Al.

P.S. I actually hope that if MAM1 takes off well, it'll [slightly?] reduce the volume of MCM1 work out there at any given time which, with luck, might make the sending of missed-deadline retries a lot easier! (Yes, I know - I'm repeating myself; it's a characteristic of old age, I'm told...)

[Edited to slightly rephrase the reference to HR classes and cache blocking.]

I see the missed deadline work units appear seemingly in batches. It kind of reminds me of the old data processing routines on mainframes involving batch processing. The order and processing of various different kinds of batches was a well orchestrated dance so various subsystems (memory, I/O, cpu usage, etc) did not get overloaded to the detriment of other batches.

Cheers

I've got a good amount of MCM and ARP WUs. Hopefully we have a good weekend of nice flows.

And it's all goodnight again:

30-Mar-2025 08:42:56 [World Community Grid] Scheduler request completed
30-Mar-2025 08:42:56 [World Community Grid] Server can't open database 
30-Mar-2025 08:42:56 [World Community Grid] Project requested delay of 3600 seconds