micro benchmark - build

[yishayahu]

We are trying to build and run the micro benchmark.
We get an error in the compilation - no "select_row" is defined for ordered_index.
Indeed, looking at the "DB_oindex.hh" we see that this function implementation is under "if 0".
Would you please explain to us if and how to fix that?
Thanks,
Shaya

[wqian94]

For the purpose of testing our timestamp splitting via split versions implementation, we replaced select_row with select_split_row. If you do not intend to use this feature, select_row should work just fine if you uncomment it.

[yishayahu]

we tried to uncomment select_row, it is not working for us.
Thanks

[wqian94]

Is this after uncommenting both select_row overrides? One of them calls the other, so both should be uncommented.

If you're using commit 013f5e, these are lines 118-132 and 150-195.

[yishayahu]

yes, we tried to uncomment both of them, and that's the commit we are using.

[wqian94]

Can you please send the error trace that you are seeing?

[yishayahu]

benchmark/DB_oindex.hh:129:66: error: could not convert ‘{true, false, 0, }’ from ‘’ to ‘bench::ordered_index<ubench::MasstreeIntKey, eight_version_row, db_params::db_swiss_params>::sel_return_type {aka std::tuple<bool, bool, long unsigned int, const eight_version_row*>}’
return {true, false, 0, UniRecordAccessor(nullptr)};

[wqian94]

Ah -- do you have UniRecordAccessor defined for eight_version_row? UniRecordAccessor should be auto-coerced into the appropriate pointer type.

[yishayahu]

no, how can I define it?

[wqian94]

It's a little roundabout, but if you look at lines 4-236 in benchmark/tpcc_split_params_default.hh, that's an example of the specialized template classes for warehouse_value from TPC-C. In particular, the UniRecordAccessor and SplitRecordAccessor classes are essentially the same for this case.

If you're looking to use the timestamp splitting feature, you'll want to look at benchmark/tpcc_split_params_ts.hh instead.