In pr #8476, all memory usage of a process is recorded in the process mem tracker,
and all memory usage of a query is recorded in the query mem tracker,
and it is still necessary to manually call `transfer to` to track the cached memory size.
We hope to separate out more detailed memory usage based on Hook TCMalloc new/delete + TLS mem tracker.
In this pr, the more detailed mem tracker is switched to TLS, which automatically and accurately
counts more detailed memory usage than before.
This is only a temporary fix its performance is not ideal. Finally,
we need to reconstruct the functions of `stddev` and delete the interface of `insert_to_null_default ()`.
Add a new column-type to speed up the approximation of quantiles.
1. The new column-type is named `quantile_state` with fixed aggregation function `quantile_union`, which stores the intermediate results of pre-aggregated approximation calculations for quantiles.
2. support pre-aggregation of new column-type and quantile_state related functions.
This feature is propsoed in [DSIP-1](https://cwiki.apache.org/confluence/display/DORIS/DSIP-001%3A+Java+UDF).
This PR support fixed-length input and output Java UDF. Phase I in DIP-1 is done after this PR.
To support Java UDF effeciently, I use no data copy in JNI call and all compute operations are off-heap in Java.
To achieve that, I use a UdfExecutor instead.
For users, a UDF class must have a public evaluate method.
Currently, the compiled output of BE mainly consists of two binaries:
palo_be and meta_tool, which are both around 1.6G in size.
However, the debug information is only needed for debugging purposes.
So I separate the debug info from binaries.
After BE is built, the debug info file will be saved in `be/lib/debug_info/` dir.
`palo_be` and `meta_tool`'s size decrease to about 100MB
This is optional, and default is disabled.
To enable it, use:
`STRIP_DEBUG_INFO=ON sh build.sh`
WindowFunctionLastData::add should keep the last value,
but current implementation keeps the first one.
Obviously, this code is copied from WindowFunctionFirstData::add.
Early Design Documentation: https://shimo.im/docs/DT6JXDRkdTvdyV3G
Implement a new way of memory statistics based on TCMalloc New/Delete Hook,
MemTracker and TLS, and it is expected that all memory new/delete/malloc/free
of the BE process can be counted.
1.
The methods in the IndexChannel are called back in the RpcClosure in the NodeChannel.
However, this callback may occur after the whole task is finished (e.g. due to network latency),
and by that time the IndexChannel may have been destructured, so we should not call
the IndexChannel methods anymore, otherwise the BE will crash.
Therefore, we use the `_is_closed` variable and `_closed_lock` to ensure that the RPC callback function
will not call the IndexChannel's method after the NodeChannel is closed.
2.
Do not add IndexChannel to the ObjectPool.
Because when deconstruct IndexChannel, it may call the deconstruction of NodeChannel.
And the deconstruction of NodeChannel maybe time consuming(wait rpc finished).
But the ObjectPool will hold a SpinLock to destroy the objects, so it may cause CPU busy.
The current BloomFilter runs vectorization predicate evaluate, but `evaluate_vec` interface is not implemented, so the RuntimeFilter does not play a role after it is pushed down to the storage layer.
And BF predicate computation cannot be automatically vectorized, thus making BloomFilter run short-circuit logic.
For SSB Q2.1，`enable_storage_vectorization = true;`
```
test before impl:
- Total: 36s164ms
- RowsVectorPredFiltered: 0
- RealRuntimeFilterType: bloomfilter
- HasPushDownToEngine: true
test after impl:
- Total: 2s345ms
- RowsVectorPredFiltered: 595.247102M (595247102)
- RealRuntimeFilterType: bloomfilter
- HasPushDownToEngine: true
```
Fix a bug when query bitmap return wrong result, even the simplest query.
Such as
```
CREATE TABLE `pv_bitmap_fix2` (
`dt` int(11) NULL COMMENT "",
`page` varchar(10) NULL COMMENT "",
`user_id_bitmap` bitmap BITMAP_UNION NULL COMMENT ""
) ENGINE=OLAP
AGGREGATE KEY(`dt`, `page`)
COMMENT "OLAP" DISTRIBUTED BY HASH(`dt`) BUCKETS 2
PROPERTIES (
"replication_allocation" = "tag.location.default: 1",
"in_memory" = "false",
"storage_format" = "V2"
)
Insert any hundreds of rows of data
select count(distinct user_id_bitmap) from pv_bitmap_fix2
the result is wrong
```
This is a bug of vectorization of storage layer.
Due to unlimited queue in OlapScanNode and NodeChannel, memory usage can be
very large for reading and writing large table, e.g 'insert into tableB select * from tableA'.
