sync operates on a spoke and hub model (parent server with children)
each child needs to initiated with a copy of the parent server's DB
this was done through database dumping
an alternate option to consider is to copy the mysql files instead of importing/exporting the DB, in order to save import/export time
any way to not start out with clones and have some kind of merge done?
short answer is no since the records would need to be reconciled
the key to doing that successfully is to be able to match up UUIDs, which is a non-trivial task
registration is done between the child to parent and parent to children
parent server and child servers have a static IP
it may be possible to use a dynamic DNS (DDNS) service (e.g. no-ip.org, dyndns.com) to get around this
sync process always runs on the child; configured to contact the parent server on a regular basis
can we use the parent server as the operational server?
yes, this is how it is currently used in PIH Rwanda
is there a performance impact?
none observed so far
configurable settings
the amount of time when the child contacts the parent
timeout for when the child should stop contacting the parent
maximum number of tries for the child to reach the parent
whether sync'ed data is compressed or not (which is run through the openmrs configuration which can optionally be gzip)
the class of data to be sync'ed
at the technical level, it is sync'ing at the level of db tables
not able to sync subsets of patients, but this is a feature we'd like to implement in the future
prerequisites
did we need to beef up our servers for the sync?
no, we have our servers running on off the shelf laptops with 3GB of RAM
had issues with the servers not being on proper UPS (we used laptops)
mysql databases can get corrupted if power is lost in the middle of a transaction
what about network firewall considerations?
this is done via HTTP post, so either port 80 (HTTP) or 443 (HTTPS)
bandwidth
our sites are on VSAT and performance has been fine even with all of the outages that we have. sometimes the bandwidth is 1kbps or less.
how long do we need to have something up and running in order for the sync'ing to work?
dependent on bandwidth, but from our experience, sending 200 records at a time, it takes about 5 minutes to sync
this can even be sync'ed via USB key
and there won't be duplicates even if it is manually sync'ed by USB and afterwards, the network is restored
we will be piloting this on GPRS modems; the issue is not bandwidth, but rather having fixed IP addresses on the children (that are using the modem)
however, this may be overcome by using a DDNS service (see above)
functional questions
still able to do updates on both the parent and the child?_
yes
what is the format of what is being sent?
xml (serialized java objects)
can the XML be consumed by other services?
probably possible
is it possible to have grandparents? i.e. multiple layers of syncing between children and the parent?
theoretically it could be possible
sync was built to be a child to parent relationship, where each node will communicate to a parent (if it exists) and its children (if they exist); children don't know about each other.
is there any thought about doing a peer-to-peer relationship rather than a hierarchical relationship?
this might be something to consider in the future
(a concern was raised by an audience member): not sure if this should be perceived as a solution for a national reporting system built off of (child) districts and (grandchild) clinics because of data access/privacy concerns; DIHS should be used instead
filtering might help to alleviate this challenge
complex obs files able to be sync'ed?
may not be possible to do at the current time since it would require so much bandwidth to do
2 vs. 1 way sync
it is possible to use the sync module as an info path alternative (in a one directional way)
similar/related efforts
run a sync on an android device
the android device has a full DB copy
it is not made to handle conflicts
questions to think about:
is there any opportunity to combine the efforts through refactoring?
because there has been trouble to get hibernate to run on android, is there any way to remove that dependency from the existing sync module?