[1839] in Kerberos-V5-bugs
Use of gdbm as database.... (Linux)
daemon@ATHENA.MIT.EDU (epeisach@MIT.EDU)
Mon Mar 18 19:03:45 1996
From: epeisach@MIT.EDU
Date: Mon, 18 Mar 1996 19:03:41 -0500
To: krb5-bugs@MIT.EDU
For linux, gdbm is the chosen default database for better or worse. The
bad part, if I am reading the code correctly is that gdbm does its own
database locking at the time of the database open. For multiple
processes, this can be a problem. The database can be opened by many
readers at one time, or by a single writer. As I see it we are sort of
screwed as you cannot run more than one server or process if you plan to
modify the database (i.e. krb5dc, kadmin, kdb5_edit - all losing combos)
I see two options:
a) The database code will open/close the database for every transaction
- a real hit in performance for those read-only servers.
b) Some sort of IPC to tell other servers to close the files - This is
blech - I wouldn't trust it...
c) Tell users in the installation manual that gdbm is not an option at
this time and to configure with the berk_db code...
d) Autodetect gdbm and automagically configure in berk_db...
Personally, I prefer (d)
Ezra
