I finished my first cut at handling an IP address change.

The getopt() switch() statement in main() will now handle a case 'c': for a "-c" option, which will flag Dynamic IP (or DNS) checking.  This will cause a routine called DynamicChecking() to be fork()'ed, which performs the indicated logic of periodically scanning the BRIDGE array to see if the results of a gethostbyname() of the host field still agrees with the addr field (I.E., the in_addr struct).  DynamicChecking() is not long; 40 lines of code, yet most of it is made up of various status messages, depending on Verbosity.

For interested parties, the alpha version may grabbed via anonymous NFT on VENTI2::BRIDGE.C.  Note that this is not fully tested yet and currently has conditionals for Xcode's odd debugging environment.  In particular, I had two thoughts about the code. 

  1. If a SIGHUP comes in while the main loop is handling a message, does the right thing happen or does an explosion occur?
  2. Does the bind() need to get redone after the triggered read_conf()?  I guess only if the port changes, but read_conf() doesn't touch that.

On 4/26/22 8:05 PM, Thomas DeBellis wrote:

I'll have a look some more, then; I haven't crashed any of my Unix boxes in a while. 

I finally found the pernicious bug I that introduced into TTYSRV when I was updating it to make my life in the Access Control Job easier.  In short, you will sometimes get what deserve if you dereference an index register that you naively thought somebody else set up for you.  If you happen to do this in Exec mode, then you will be ever-so-more richly rewarded.

What a rabbit hole...  Still, now PANDA can remark on who is looking over who's shoulder.

Nice to finally come up for air, anyway.

On 4/26/22 6:47 PM, Johnny Billquist wrote:

Hi.

No. You are right. That could definitely be done. I've just been to lazy to ever start polishing things...

  Johnny

On 2022-04-26 23:03, Thomas DeBellis wrote:

Johnny,

Funny you should say that.

As soon as you responded below, I had a look at bridge.c.   Basically it looks like I'd just need to do a fork().  That child process would wake up ever so often and do a gethostbyname() on each host field in the BRIDGE structure and compare that IP address with what was in the corresponding addr field.  Any difference would mean it would send the parent a SIGHUP to reparse the bridge.conf file, which would rebuild the BRIDGE structure with the correct IP addresses.

If we wanted to keep the accumulated totals (rcount, zcount, Etc.) correct for SIGUSR1, then maybe those could be saved before the SIGHUP and added back after this was complete.

That doesn't sound too awful, unless I'm missing something (haven't really gotten into that code for a year or two).
--T
On 4/26/22 3:45 PM, Johnny Billquist wrote:

Well. The thing is, this was/is a listening socket. So there is no reconnecting. It's just about binding.
On 2022-04-26 19:40, Paul Koning wrote:

PyDECnet rechecks the name resolution every hour.  It doesn't close existing connections, but if a node moves and its DNS entry is updated to reflect that, PyDECnet will find it again and reconnect within an hour.

On Apr 26, 2022, at 1:22 PM, Johnny Billquist <bqt@softjar.se> wrote:

The problem is that dynamic dns don't necessarily solve any problems. The address is bound at start, and if the resolution in DNS changes at some later point, that won't get reflected unless pydecnet would redo the name lookup, close and reopen the listening socket bound to the new address.