harvey/sys/man/8/replica

.TH REPLICA 8
.SH NAME
applychanges, applylog, compactdb, updatedb \- simple client-server replica management
.SH SYNOPSIS
.B replica/compactdb
.I db
.br
.B replica/updatedb
[
.B -cl
]
[
.B -p
.I proto
]
[
.B -r
.I root
]
[
.B -t
.I now
.I n
]
[
.B -u
.I uid
]
[
.B -x
.I path
] ...
.I db
.br
.B replica/applylog
[
.B -cnsuv
]
.I clientdb
.I clientroot
.I serverroot
[
.I path ...
]
.br
.B replica/applychanges
[
.B -nuv
]
[
.B -p
.I proto
]
[
.B -x
.I path
] ...
.I clientdb
.I clientroot
.I serverroot
[
.I path ...
]
.SH DESCRIPTION
These four tools collectively provide simple log-based
client-server replica management.
The shell scripts described in
.IR replica (1)
provide a more polished interface.
.PP
Both client and server maintain textual databases of file system
metadata.  Each line is of the form
.sp 0.5
\h'0.25i'
.I path
.I mode
.I uid
.I gid
.I mtime
.I length
.sp 0.5
Later entries for a
.I path
supersede previous ones.
A line with the string
.B REMOVED
in the
.I mode
field annuls all previous entries for that
.IR path .
The entries in a file are typically kept sorted by
.I path
but need not be.
These properties facilitate updating the database atomically
by appending to it.
.I Compactdb
reads in a database and writes out an equivalent one,
sorted by path and without outdated or annulled records.
.PP
A replica is further described on the server by a textual
log listing creation and deletion of files and changes
to file contents and metadata.
Each line is of the form:
.sp 0.5
\h'0.25i'
.I time
.I gen
.I verb
.I path
.I serverpath
.I mode
.I uid
.I gid
.I mtime
.I length
.sp 0.5
The
.I time
and
.I gen
fields are both decimal numbers, providing an ordering 
for log entries so that incremental tools need not process
the whole log each time they are run.
The
.IR verb ,
a single character,
describes the event:
addition of a file
.RB ( a ),
deletion of a file
.RB ( d ),
a change to a file's contents
.RB ( c ),
or a change to a file's metadata
.RB ( m ).
.I Path
is the file path on the client;
.I serverpath
the path on the server (these are different when the
optional fifth field in a proto file line is given;
see
.IR proto (2)).
.IR Mode ,
.IR uid ,
.IR gid ,
and
.I mtime
are the files metadata as in the
.B Dir
structure (see
.IR stat (5)).
For deletion events, the metadata is that of the deleted file.
For other events, the metadata is that after the event.
.PP
.I Updatedb
scans the file system rooted at
.I root
for changes not present in
.IR db ,
noting them by appending new entries to the database
and by writing log events to standard output.
The
.B -c
option causes
.I updatedb
to consider only file and metadata changes, ignoring file additions and deletions.
By default, the log events have
.I time
set to the current system time
and use incrementing 
.I gen
numbers starting at 0.
The
.B -t
option can be used to specify a different time and starting number.
If the
.B -u
option is given, all database entries and log events will use
.I uid
rather than the actual uids.
The
.B -x
option (which may be specified multiple times) excludes the named path
and all its children from the scan.
If the
.B -l
option is given, the database is not changed and the
.I time
and
.I gen
fields are omitted from the log events;
the resulting output is intended to be a human-readable
summary of file system activity since the last scan.
.PP
.I Applylog
is used to propagate changes from server to client.
It applies the changes listed in a log
(read from standard input)
to the file system rooted at
.IR clientroot ,
copying files when necessary from the file system rooted at
.IR serverroot .
By default,
.I applylog
does not attempt to set the uid on files; the
.B -u
flag enables this.
.I Applylog
will not overwrite local changes made to replicated files.
When it detects such conflicts, by default it prints an error describing
the conflict and takes no action.
If the
.B -c
flag is given, 
.I applylog
still takes no action, but does so silently and will not
report the conflicts in the future.
(The conflict is resolved in favor of the client.)
If the
.B -s
flag is given,
.I applylog
overwrites the local changes.
(The conflict is resolved in favor of the server.)
.PP
.I Applychanges
is, in some sense, the opposite of
.IR applylog ;
it scans the client file system for changes, and applies
those changes to the server file system.
.I Applychanges
will not overwrite remote changes made to replicated files.
For example, if a file is copied from server to client and subsequently
changed on both server and client,
.I applychanges
will not copy the client's new version to the server, because
the server also has a new version.
.I Applychanges
and
.I applylog
detect the same conflicts; to resolve conflicts reported by
.IR applychanges ,
invoke
.I applylog
with the
.B -c
or
.B -s 
flags.
.SH EXAMPLE
One might
keep a client kfs file system up-to-date
against a server file system using these tools.
First, connect to a CPU server with a high-speed
network connection to the file server and scan
the server file system, updating the server database and log:
.EX
    repl=$home/lib/replica
    proto=/sys/lib/sysconfig/proto/portproto
    db=$repl/srv.portproto.db
    log=$repl/srv.portproto.log

    9fs $fs
    replica/updatedb -p $proto -r /n/$fs -x $repl $db >>$log
    replica/compactdb $db >/tmp/a && mv /tmp/a $db
.EE
.PP
Then, update the client file system:
.EX
    repl=$home/lib/replica
    db=$repl/cli.portproto.db
    log=$repl/srv.portproto.log

    9fs $fs
    9fs kfs
    replica/applylog $db /n/kfs /n/$fs <$log
    replica/compactdb $db >/tmp/a && mv /tmp/a $db
.EE
.PP
The
.B $repl
directory is excluded from the sync so that multiple
clients can each have their own local database.
The shell scripts in
.B /rc/bin/replica
are essentially a further development of this example.
.PP
The Plan 9 distribution update program
operates similarly, but omits the first scan;
it is assumed that the Plan 9 developers run
scans manually when the distribution
file system changes.
The manual page
.IR replica (1)
describes this in full.
.SH SEE ALSO
.IR replica (1)
.SH BUGS
These tools assume that 
.I mtime
combined with
.I length
is a good indicator of changes to a file's contents.