| Commit message (Collapse) | Author | Age | Files | Lines |
|---|
| |
|
|
| |
changes, re-queue failed directory sync packets, fix account_id index in queue creation, be more selective about updating hubloc_connected (only when we're talking to the associated site). But we still need a way to flag dead accounts and these mechanisms won't cut it, because it isn't a requirement that channels communicate with anybody, except for contacting the directory when the account is created. I think we need to make every channel ping the directory once a month. Then we can find those that have not done so (after syncing with other directories). We can't have the directory just flag dead accounts as this would create a way for an anti-social attacker to become a directory server and mark all the accounts dead.
|
| | |
|
| | |
|
| | |
|
| | |
|
| |
|
|
|
|
|
|
|
| |
possibly recurse and blow up the matrix. Hard to say. Do you feel lucky? Well do ya' ... punk? Rule #1 - don't mess with anything unless it's blowing up the matrix. If it doesn't blow up the matrix, but doesn't work, just let it go and let's figure out what it is doing and what it isn't doing.
The flow is as follows:
Once a day go out to all the directory servers besides yourself and grab a list of updates. This happens in the poller. If we've never seen them before add them to the updates table. The poller also looks to see if we're a directory server and have updates that haven't yet been processed. It calls onedirsync.php to process each one. If we contact the channel to update and don't find anything (we're just doing a basic zot_finger), set a ud_last timestamp. If this is set we will only try once a day for seven days. Then we stop trying to update.
This will probably cause a spike the first time through because you haven't seen any updates before, but we spread out the load over your delivery interval.
|
| | |
|
| |
|
|
| |
nothing else
|
| | |
|
| | |
|
| | |
|
| | |
|
| |
|
|
| |
forward again. Put back xtag for efficient keyword searching, use it in conjunction with xprof['keywords'] to avoid a SQL triple join and group_concat (which simply won't scale on a directory server), figured out how best to work timezone corrected birthdays into the protocol. The directory can even provide an age if one was provided to it. Both of these things need some more work, but we have the mechanisms and a plan how to do it right instead of "how the !@#$ are we going to do pull this off?"
|
| | |
|
| | |
|
|
|
standalone, need to put some basic profile info in discovery page and direct to local directory master
|
