Is this library has high consistency? about monstache

[rwynn]

Documents are batched for indexing until one of your options fires: elasticsearch-max-docs, elasticsearch-max-bytes, or elasticsearch-max-seconds. These options trigger a flush at which point a _bulk API request is made with those documents. This request could fail (some or all of the documents not indexed). If it fails and you have set elasticsearch-retry-seconds to a value > 0, that failed request will occur again after the given amount of seconds. If that request fails a 2nd time, the request (including the documents sent) are logged as an error.

Currently, monstache will continue processing when the situation above occurs. That is, it continues processing and sending batches of documents coming off the oplog after an error. I am considering adding a fail-fast option which when set to true would exit the program on the first _batch send error.

Note: monstache includes options replay and replay-from-timestamp. When the former is true monstache replays all events from the beginning of the oplog. When the latter is supplied monstache replays events starting after the timestamp.

Does that answer your question? Would you be interested in the fail-fast option?

[asaf]

Ryan,
thanks for the detailed answer,

I think it makes sense to add the fail-fast option to avoid repetitive errors when there is a constant failure on the ES service,

Regarding the replay, so in the case of missing docs due to errors, if I run monstache with the replay option once in a while, will it re-index the missing docs? will it update/skip existing docs or should the indices get wiped manually first?

Thanks

[rwynn]

@asaf
when you run with replay it will re-attempt to index the docs, so docs that failed previously may succeed. You can delete the indices in Elasticsearch but it's not necessary, the docs that were previously successful with be re-indexed with the same data.

A scenario that would not be fixed by replay would be a situation where you have 2 documents in a collection with a property of different types. e.g. { age: 1} and {age: "2"}. The first doc will establish the type in ES of age as long. Then the 2nd document would always fail because a string is sent. This type of error would not be fixed by replay. Replay would fix an issue like ES becoming temporarily unavailable.

[asaf]

That's an accepted behavior :-)

[rwynn]

@asaf,
In the next release (shortly) there will be an option to specify a list of collections that you would like to copy directly from mongodb to elasticsearch. This is different than the current situation which allows you to replay and then tail the oplog. The direct sync option will read the collections themselves and not go through the oplog. Since the oplog may or may not contain your entire dataset (it's a capped collection after all), this direct copy option can be used instead. The option to exit monstache once the direct copy is complete will be added along with it (so it can be used in cron jobs for example).