You can clone with
As pointed out on the mailing list there are performance issues regarding the integration of the PB client/pipe/mapred/search (1). Essentially mapred/search can take orders of magnitude longer via the PB client because of issues with pipe/search integration. Until this is fixed there are two work arounds:
Use legacy mapreduce (2)
If writing an Erlang application you can still use pipe for mapreduce but make rpc calls to the search:mapred function. This function is still hardcoded to use the legacy system and therefore won't be affected if the general mapreduce system is using pipe.
@rzezeski We should talk about native PB endpoints for Search (and other apps).
This impacts a customer's use of search: zd #1189
Still valid? We now have a native PB endpoint for Search.