**Histogram statistics**
Currently doris collects statistics, but no histogram data, and by default the optimizer assumes that the different values of the columns are evenly distributed. This calculation can be problematic when the data distribution is skewed. So this pr implements the collection of histogram statistics.
For columns containing data skew columns (columns with unevenly distributed data in the column), histogram statistics enable the optimizer to generate more accurate estimates of cardinality for filtering or join predicates involving these columns, resulting in a more precise execution plan.
The optimization of the execution plan by histogram is mainly in two aspects: the selection of where condition and the selection of join order. The selection principle of the where condition is relatively simple: the histogram is used to calculate the selection rate of each predicate, and the filter with higher selection rate is preferred.
The selection of join order is based on the estimation of the number of rows in the join result. In the case of uneven data distribution in the join condition columns, histogram can greatly improve the accuracy of the prediction of the number of rows in the join result. At the same time, if the number of rows of a bucket in one of the columns is 0, you can mark it and directly skip the bucket in the subsequent join process to improve efficiency.
---
Histogram statistics are mainly collected by the histogram aggregation function, which is used as follows:
**Syntax**
```SQL
histogram(expr)
```
> The histogram function is used to describe the distribution of the data. It uses an "equal height" bucking strategy, and divides the data into buckets according to the value of the data. It describes each bucket with some simple data, such as the number of values that fall in the bucket. It is mainly used by the optimizer to estimate the range query.
**example**
```
MySQL [test]> select histogram(login_time) from dev_table;
+------------------------------------------------------------------------------------------------------------------------------+
| histogram(`login_time`) |
+------------------------------------------------------------------------------------------------------------------------------+
| {"bucket_size":5,"buckets":[{"lower":"2022-09-21 17:30:29","upper":"2022-09-21 22:30:29","count":9,"pre_sum":0,"ndv":1},...]}|
+------------------------------------------------------------------------------------------------------------------------------+
```
**description**
```JSON
{
"bucket_size": 5,
"buckets": [
{
"lower": "2022-09-21 17:30:29",
"upper": "2022-09-21 22:30:29",
"count": 9,
"pre_sum": 0,
"ndv": 1
},
{
"lower": "2022-09-22 17:30:29",
"upper": "2022-09-22 22:30:29",
"count": 10,
"pre_sum": 9,
"ndv": 1
},
{
"lower": "2022-09-23 17:30:29",
"upper": "2022-09-23 22:30:29",
"count": 9,
"pre_sum": 19,
"ndv": 1
},
{
"lower": "2022-09-24 17:30:29",
"upper": "2022-09-24 22:30:29",
"count": 9,
"pre_sum": 28,
"ndv": 1
},
{
"lower": "2022-09-25 17:30:29",
"upper": "2022-09-25 22:30:29",
"count": 9,
"pre_sum": 37,
"ndv": 1
}
]
}
```
TODO:
- histogram func supports parameter and sample statistics (It's got another pr)
- use histogram statistics
- add p0 regression
The segment group is useless in current codebase, remove all the related code inside Doris. As for the related protobuf code, use reserved flag to prevent any future user from using that field.
Currently, newly created segment could be chosen to be compaction
candidate, which is prone to bugs and segment file open failures. We
should skip last (maybe active) segment while doing segcompaction.
1.remove quick_compaction's rowset pick policy, call cu compaction when trigger
quick compaction
2. skip tablet's compaction task when compaction score is too small
Co-authored-by: yixiutt <yixiu@selectdb.com>
1. change jsonb_extract_string behavior: convert to string instead of NULL if the type of json path is not string
2. move jsonb tutorial doc to JSONB data type
#13195 left some unresolved issues. One of them is that some BE unit tests fail.
This PR fixes this issue. Now, we can run the command ./run-be-ut.sh --run successfully on macOS.
PR https://github.com/apache/doris/pull/13917 has supported lazy read for non-predicate columns in ParquetReader,
but can't trigger lazy read when predicate columns are partition or missing columns.
This PR support such case, and fill partition and missing columns in `FileReader`.
Read predicate columns firstly, and use VExprContext(push-down predicates)
to generate the select vector, which is then applied to read the non-predicate columns.
The data in non-predicate columns may be skipped by select vector, so the value-decode-time can be reduced.
If a whole page can be skipped, the decompress-time can also be reduced.
mem tracker can be logically divided into 4 layers: 1)process 2)type 3)query/load/compation task etc. 4)exec node etc.
type includes
enum Type {
GLOBAL = 0, // Life cycle is the same as the process, e.g. Cache and default Orphan
QUERY = 1, // Count the memory consumption of all Query tasks.
LOAD = 2, // Count the memory consumption of all Load tasks.
COMPACTION = 3, // Count the memory consumption of all Base and Cumulative tasks.
SCHEMA_CHANGE = 4, // Count the memory consumption of all SchemaChange tasks.
CLONE = 5, // Count the memory consumption of all EngineCloneTask. Note: Memory that does not contain make/release snapshots.
BATCHLOAD = 6, // Count the memory consumption of all EngineBatchLoadTask.
CONSISTENCY = 7 // Count the memory consumption of all EngineChecksumTask.
}
Object pointers are no longer saved between each layer, and the values of process and each type are periodically aggregated.
other fix:
In [fix](memtracker) Fix transmit_tracker null pointer because phamp is not thread safe #13528, I tried to separate the memory that was manually abandoned in the query from the orphan mem tracker. But in the actual test, the accuracy of this part of the memory cannot be guaranteed, so put it back to the orphan mem tracker again.
## Design
### Trigger
Every time when a rowset writer produces more than N (e.g. 10) segments, we trigger segment compaction. Note that only one segment compaction job for a single rowset at a time to ensure no recursing/queuing nightmare.
### Target Selection
We collect segments during every trigger. We skip big segments whose row num > M (e.g. 10000) coz we get little benefits from compacting them comparing our effort. Hence, we only pick the 'Longest Consecutive Small" segment group to do actual compaction.
### Compaction Process
A new thread pool is introduced to help do the job. We submit the above-mentioned 'Longest Consecutive Small" segment group to the pool. Then the worker thread does the followings:
- build a MergeIterator from the target segments
- create a new segment writer
- for each block readed from MergeIterator, the Writer append it
### SegID handling
SegID must remain consecutive after segment compaction.
If a rowset has small segments named seg_0, seg_1, seg_2, seg_3 and a big segment seg_4:
- we create a segment named "seg_0-3" to save compacted data for seg_0, seg_1, seg_2 and seg_3
- delete seg_0, seg_1, seg_2 and seg_3
- rename seg_0-3 to seg_0
- rename seg_4 to seg_1
It is worth noticing that we should wait inflight segment compaction tasks to finish before building rowset meta and committing this txn.
