1998-08-09 08:43:13 +08:00
|
|
|
Design Notes: slurpd
|
|
|
|
|
|
|
|
|
|
This new version differs significantly from previous versions:
|
|
|
|
|
|
|
|
|
|
- It uses a multithreaded, single-process model. Previous versions forked
|
|
|
|
|
a separate process for each replica. This new design should facilitate
|
|
|
|
|
porting to NT, and is more straightforward.
|
|
|
|
|
|
|
|
|
|
- Only one copy of the replication log is made. Previous versions made
|
|
|
|
|
one copy of the replication log for each replica
|
|
|
|
|
|
|
|
|
|
- This newer version is more object-oriented. Although still written in
|
|
|
|
|
ANSI C (and compilable on k&r compilers), it encapsulates within the
|
|
|
|
|
major data structures the methods used to access "private" data.
|
|
|
|
|
|
|
|
|
|
General design overview:
|
|
|
|
|
|
|
|
|
|
The main data structure in slurpd is a replication queue (struct rq).
|
|
|
|
|
The rq data structure is currently implemented as a linked list of
|
|
|
|
|
replication entries (struct re). The rq structure contains member functions
|
|
|
|
|
used to initialize, add to, remove, and return the next re struct.
|
|
|
|
|
|
|
|
|
|
In addition to the rq structure, there is one ri (replication information)
|
|
|
|
|
struct for each replica. The ri struct encapsulates all information
|
|
|
|
|
about a particular replica, e.g. hostname, port, bind dn. The single
|
|
|
|
|
public member function, ri_process, is called to begin processing
|
|
|
|
|
the replication entries in the queue.
|
|
|
|
|
|
|
|
|
|
There is also a status structure (struct st) which contains the timestamp
|
|
|
|
|
of the last successful replication operation for each replica. The
|
|
|
|
|
contents of the st struct are flushed to disk after every successful
|
|
|
|
|
operation. This disk file is read upon startup, and is used to allow
|
|
|
|
|
slapd to "pick up where it left off".
|
|
|
|
|
|
|
|
|
|
Threading notes:
|
|
|
|
|
|
|
|
|
|
The LDAP liblthread quasi-pthreads interface is used for threading. At
|
|
|
|
|
this point, machines which do not support pthreads, sun threads or lwp
|
|
|
|
|
will probably not be able to run slurpd. Given the current threading
|
|
|
|
|
method, discussed in the next paragraph, it will probably be necessary to
|
|
|
|
|
have a separate hunk of code which handles non-threaded architectures
|
|
|
|
|
(or we might just not worry about it). This needs further discussion.
|
|
|
|
|
|
|
|
|
|
Upon startup, command-line arguments and the slapd configuration file
|
|
|
|
|
are processed. One thread is started for each replica. Thread replicas,
|
|
|
|
|
when no more work exists, wait on a condition variable, and the main
|
|
|
|
|
thread's file manager routine broadcasts on this condition variable
|
|
|
|
|
when new work is added to the queue.
|
|
|
|
|
|
2001-12-08 15:38:14 +08:00
|
|
|
Additional notes:
|
|
|
|
|
See doc/devel/replication-notes.txt
|
