NetNS start issue

[svinota]

From the IRC:

< Prokto > @svinota Hi, I'm using pyroute2 to create a basic SDN type of program. I'm very thankful for the library and I'm just trying to figure out what is happening with an issue I'm having. Whenever I create IPDB objects with a NetNS object I eventually run into an issue where my python script just gets stuck and doesn't move. If I interrupt the hanging function with Control+C, I see the same error more or less. The odd thing is I can create more IPDB and < Prokto > NetNS objects without a problem even at this point but if I try to open, for example, IPDB(nl=NetNS('test1')) it just hangs there. I've tried adding a "del ipdb" to my functions that need an IPDB object but I still haven't figure out the source of the problem. I just see the amount of python processes rise until it stops. It seems like gc isn't collecting the old objects but I'm quite honestly a beginner so I'm really not sure what to look for. Sorry < Prokto > for the word-vomit here. < Prokto > Error: https://pastebin.com/SmtN7fL1 < Prokto > Example function: https://pastebin.com/NyH8yTZu < Prokto > and I'm importing "netns" from pyroute2 under another name so that shouldn't be an issue.

Looks like a regression introduced with the recent changes in the Remote class

[svinota]

An fd leak. If one passes an external RTNL API object to the IPDB constructor, IPDB doesn't close it on release(). Bad design? Bad docs?

[nickhuber]

The pastebin no longer exists so I don't know if it is the same problem but I think I am seeing the same hanging behaviour and saw it was stuck on some select.select(inputs, outputs, inputs) in pyroute2/remote/__init__.py. I saw that my program had than 1000 file descriptors so it was running into the limit of FD_SETSIZE. Maybe this could be changed to use select.poll to avoid the fd size limit incase someone already had many files open?