URL:
https://github.com/SSSD/sssd/pull/5532
Title: #5532: Handle ldap_install_tls() configuration and retrial
alexey-tikhonov commented:
"""
> **Wrt 1st patch**
I've run a battery of tests by setting only one of the communication flags per test.
I've uploaded the most prominent strace logs to the bugzilla in the attachment called
`strace battery of tests`.
(I didn't check the logs yet but ...)
Setting `LDAP_OPT_CONNECT_ASYNC` uses the read()
Setting `LDAP_OPT_NETWORK_TIMEOUT` also uses the read()
Setting `LDAP_OPT_RESTART` uses the read()
Why did we see poll() previously?
Anyway, at this point I'm totally lost wrt rationale of the 1st patch.
> **Wrt 2nd patch**
> I think we shouldn't rely on undocumented usage of `errno`, i.e. it's better
to check watchdog's context.
Do you mean watchdog_ctx structure?
Yes. Take a note: it's private. Please avoid making it public entirely, better
introduce a helper to read required info.
Would it be enough to compare the timestamp in that structure with
the actual time and see if it's below a threshold?
Not sure I understand your idea. IMO, it's better to rely on `ticks`.
"""
See the full comment at
https://github.com/SSSD/sssd/pull/5532#issuecomment-800293802