Metrics Reference¶
This page documents Xonai-specific metrics that are displayed in the history server SQL query plans and logs.
Parquet¶
These are the metrics for the Scan parquet query plan (renamed to XonParquetScan).
Metric |
Description |
||
|---|---|---|---|
init reader time total |
The time spent computing information to be used by the Parquet reader |
||
page read time total |
The time spent reading Parquet pages into memory |
||
batch read time total |
The time spent transforming Parquet pages into data batches |
||
Aggregate¶
These are the metrics for any aggregate query plan (renamed to XonHashAggregate).
Metric |
Description |
||
|---|---|---|---|
max input batch size |
The maximum number of rows of input batches |
||
max output batch size |
The maximum number of rows of output batches |
||
peak memory size total |
The peak memory size used by tasks |
||
prepare time total |
The time spent preparing to process a batch of data, such as compiling and program invocation overheads |
||
processing time total |
The time spent processing batches of data by Xonai JIT-compiled programs |
||
Exchange¶
These are the metrics for the Exchange plan (renamed to XonExchange).
Metric |
Description |
||
|---|---|---|---|
data read time total |
The time spent reading data from the shuffle |
||
data write time total |
The time spent writing data to the shuffle |
||
repartition time total |
The time spent repartitioning data |
||
repartition spill size |
The estimated size of the data spilled to the disk used to repartition data |
||
repartition peak memory size total |
The estimated peak memory size used to repartition data |
||
Note
If there is a supported aggregate operation before an Exchange operation, repartition time and resources should be very low as this is performed during the aggregate and data only needs to be copied to the shuffle.