Comment 11 for bug 1551351

Revision history for this message
Stefan Bader (smb) wrote :

@Dough, yes, so basically did not renew but obtained a new lease (which sensibly is the same IP as before). And while the dhcp server still wonders why a client ask for some IP _again_, it also silently marks the lease time as extended for another period. And since the old clients do not die (but play dead), you accumulate a new copy every time you run it (including the initial one).

@Seth, thanks for the link. Reading through it I think it is not only potentially but very much the problem we got. The upload of isc-dhcp that started the breakage was very likely the one "no change rebuild" that was done after upgrading to bind9(.19).

Sounds like one needs to manually twiddle the build back so isc-dhcp gets a bind9 lib that has threads and epoll disabled (as a work-around). Not sure how quickly that can be done. This may still have problems on terminating the process, so maybe needs adaptation of ifdown hooks as well. But that needs to be seen. First goal is to get dhclient/-server back working.