DOCS-814 journaling

[bgrabar]

Ready to Merge.

[tychoish]

journal need not be literalized.

journaling also ensures that mongodb is crash resistent: without a journal, if mongodb exits unexpectedly, then operators must assume that the data are in an inconsistent state and should resync from a clean secondary.

If we don't make this clear, it's possible that people won't respect or value the importance of journaling.

[bgrabar]

Crash resistent or crash resilient?

What operators?

Are you saying that if journaling is enabled and the primary in a replica set crashes, that the secondaries don't need to resync from a clean secondary?

[tychoish]

resilient.

operators = administrations/users (this is, admittedly a somewhat arcane and use of the term, sorry for the confusion)

the longer story is:

• without journaling, if you shutdown uncleanly (i.e. by sending kill -9 mongod, or if it encounters an error and bails out, or there's power loss) then the data is almost certainly corrupt in some way. So you either have to run repair (which just throws away invalid BSON in the database,) or you have to resync from a clean member of the set (copy the data or just use initial sync) to ensure that the data is coherent.
• with journaling, if mongod stops, it can recover everything that it wrote to the journal (which is everything less the last 100ms at most data (by default)) and the data files will be in a consistent state after it finishes playing back the journal, without need for resync (unless, of course, the secondary has fallen off the back edge of the oplog, which is an unrelated issue that doesn't need to be documented here...

[tychoish]

omit read ahead.

[epc]

c/resistent/resistant