Client assignment can miss a (small) number of docs #344
Labels
bug
Something's wrong
:Load Driver
Changes that affect the core of the load driver such as scheduling, the measurement approach etc.
Milestone
If the total document count is not a multiple of the number of clients, Rally currently misses a few documents. This will not affect the correctness of a benchmark as the number of documents should usually be very high (in the millions) but it is incorrect nevertheless.
Consider, the following scenario: We have 16000 documents among 12 clients. In that case, each client needs to index 1333.3333 documents. Currently, each client will index 1333 documents and the last one will index 1332 (due to another bug). This gives: 1333*11 + 1332 = 15595 documents which means we miss 5 documents.
Instead, we need to assign documents per client a little bit differently: two out of three clients will need to index 1333 documents and the third one will index one document more, i.e. 1334 documents. 8 (clients) * 1333 (documents) + 4 (clients) * 1334 (documents) = 16000 documents.
This new assignment rule will ensure that we index all documents but load is still spread almost evenly across all clients.
The text was updated successfully, but these errors were encountered: