Slashdot Mirror
Laptop/Server Data Synchronization?
gbr writes "I've been trying to automatically synchronize data between a laptop and a server. When the laptop is connected to the network, I want all writes to automatically propagate across to the server. When the laptop is disconnected I want the laptop user to continue working with the local data. When the laptop is reconnected, I want the data to automatically re-sync.
The issue is, the data on the server may have changed as well, which needs to propagate back to the laptop. The data doesn't contain anything too special, no database tables etc. It does contain binary data such as executables and word processing documents. I've looked at ChironFS, Unison file sync, and drbd. ChironFS needs a manual rebuild if a connection fails, and the user needs to know which machine contains the correct data. Unison requires the user to initiate the synchronization process manually every time, and drbd is just not meant for the job at hand. How do you automatically, and invisibly to the user (except in the case of conflicts), synchronize between a laptop and a server?"
Man... You are late to the party. People have been struggling with this since the beginning of time (or so it seems). Especially database apps, where they need to work in "detached mode".
I can't give you a flat out solution, because all situations are different. But I can pass on a bit of wisdom. The most important thing for you to do is create business rules for your synchronization. If the data on the server has changed and you made changes offline, who gets priority? You will have three categories of which a file can be... Client changes get priority, Server changes get priority, and Merge files. I would stay away from the last one. If you want to keep things simple, Id go for the "Server changes get priority" approach. In short, if you took an "online" file "offline" and came back, and the server copy has changed since, your offline edits are abandoned. This way, it makes it so heavily edited files have a shorter "check out window" (even if you don't use a checkout system), and forces the person taking the file offline to coordinate with everyone else that may edit this file.
"When life gives you lemons, don't make lemonade. Make life take the lemons back!" -- Cave Johnson
http://www.coda.cs.cmu.edu/
subversion is intended for a case where you have a single data store. A modern distributed SCM -- designed for disconnected use -- would make more sense.
Personally, I play with bzr most frequently; it has a nifty Python interface (and an extensive plugin architecture) which makes it quite conducive to local scriptage. (As an example -- I have a local, filesystem-backed set of CA scripts which use bzr for transactional semantics; if a method is called which throws an exception, all filesystem changes are automatically rolled back; if a method succeeds, a commit is done to record the operation and [effectively] set a rollback point). The separation between working tree and repository is optional (by default, all working trees are also repositories -- much like BitKeeper in that respect), which makes it very handy for situations like this where you don't necessarily *have* a separate, central location which all nodes can always communicate with, and where the different trees are allowed and expected to temporarily diverge.
Deleted