* feat(joblist,runtime): this is the first part of the execution model that produces a workload that can be predicted for a given query.
- forces to UM join converter to use a value from a configuration
- replaces a constant used to control a number of outstanding requests with a value depends on column width
- modifies related Columnstore.xml values
The large side read errors mentioned there can be due to failure to
close file stream properly. Some of the data may still reside in the
file stream buffers, closing must flush it. The flush is an I/O
operation and can fail, leading to partial write and subsequent partial
read.
This patch tries to provide better diagnostics.
Co-authored-by: Sergey Zefirov <72864488+mariadb-SergeyZefirov@users.noreply.github.com>
This patch:
1. Handles corner case when the bucket exceeded the memory limit, but we cannot redistribute the data in this bucket into new buckets based on a hash algorithm, because the rows have the same values.
2. Adds force option for disk join step.
3. Add a option to contol the depth of the partition tree.
This is attempt to make some part of the code more stable.
For some reason we can get a spurious nullptr for boost::shared_ptr
which cause an assert and abort.
EM and PP are most resource-hungry runtimes.
The merge enables to control their cummulative
resource consumption, thread allocation + enables
zero-copy data exchange b/w local EM and PP facilities.
This patch changes the logic of the `receiveMultiPrimitiveMessages`
function in the following way:
1. We have only one aggregation thread which reads the data from Queue (which is populated
by messages from BPPs).
2. Processing of the received `bytestream vector` could be in parallel depends on the
type of `TupleBPS` operation (join, fe2, ...) and actual thread pool workload.
The motivation is to eliminate some amount of context switches.
* Adds CompressInterfaceLZ4 which uses LZ4 API for compress/uncompress.
* Adds CMake machinery to search LZ4 on running host.
* All methods which use static data and do not modify any internal data - become `static`,
so we can use them without creation of the specific object. This is possible, because
the header specification has not been modified. We still use 2 sections in header, first
one with file meta data, the second one with pointers for compressed chunks.
* Methods `compress`, `uncompress`, `maxCompressedSize`, `getUncompressedSize` - become
pure virtual, so we can override them for the other compression algos.
* Adds method `getChunkMagicNumber`, so we can verify chunk magic number
for each compression algo.
* Renames "s/IDBCompressInterface/CompressInterface/g" according to requirement.
* Introduce multigeneration aggregation
* Do not save unused part of RGDatas to disk
* Add IO error explanation (strerror)
* Reduce memory usage while aggregating
* introduce in-memory generations to better memory utilization
* Try to limit the qty of buckets at a low limit
* Refactor disk aggregation a bit
* pass calculated hash into RowAggregation
* try to keep some RGData with free space in memory
* do not dump more than half of rowgroups to disk if generations are
allowed, instead start a new generation
* for each thread shift the first processed bucket at each iteration,
so the generations start more evenly
* Unify temp data location
* Explicitly create temp subdirectories
whether disk aggregation/join are enabled or not
After creating and populating tables with CHAR(5) case insensitive columns,
in a set of consequent joins like:
select * from t1, t2 where t1.c1=t2.c1;
select * from t1, t2 where t1.c1=t2.c2;
select * from t1, t2 where t1.c2=t2.c1;
select * from t1, t2 where t1.c2=t2.c2;
only the first join worked reliably case insensitively.
Removing the remaining pieces of the code that used order_swap() to compare
short CHAR columns, and using Charset::strnncollsp() instead.
This fixes the issue.
