We create a new segment format for BetaRowset. New format merge
data file and index file into one file. And we create a new format
for short key index. In origin code index is stored in format like
RowCusor which is not efficient to compare. Now we encode multiple
column into binary, and we assure that this binary is sorted same
with the key columns.
When creating table with OLAP engine, use can specify multi parition columns.
eg:
PARTITION BY RANGE(`date`, `id`)
(
PARTITION `p201701_1000` VALUES LESS THAN ("2017-02-01", "1000"),
PARTITION `p201702_2000` VALUES LESS THAN ("2017-03-01", "2000"),
PARTITION `p201703_all` VALUES LESS THAN ("2017-04-01")
)
Notice that load by hadoop cluster does not support multi parition column table.
The TabletQuorumFailedException will be thrown in commitTxn while the success replica num of tablet is less then quorom replica num.
The Hadoop load does not handle this exception because the push task will retry it later.
The streaming broker, insert, stream and mini load will catch this exception and abort the txn after that.
Doris support deploy multi BE on one host. So when allocating BE for replicas of
a tablet, we should select different host. But there is a bug in tablet scheduler
that same host may be selected for one tablet. This patch will fix this problem.
There are some places related to this problem:
1. Create Table
There is no bug in Create Table process.
2. Tablet Scheduler
Fixed when selecting BE for REPLICA_MISSING and REPLICA_RELOCATING.
Fixed when balance the tablet.
3. Colocate Table Balancer
Fixed when selecting BE for repairing colocate backend sequence.
Not fix in colocate group balance. Leave it to colocate repairing.
4. Tablet report
Tablet report may add replica to catalog. But I did not check the host here,
Tablet Scheduler will fix it.
Currently, we have Field and ColumnSchema to access column data in a
row. These two classes are mostly the same. So we should unify these to
one class. Now, Field has offset information, which is an row attribute,
so we remove offset in Field.
RowCursor now has some logic which belong to Schema, so in this patch I
add Schema attribute to RowCursor to make RowCursor simple. After this
change, only Schema will handle Field/ColumnSchema.
I extract some logic from RowCursor to be/src/olap/row.h, then we can
use same logic to handle different types of row. Each type of row has
same function that to get Cell of this row. A cell represent a column
content with a null indicator.
The catch statement cancel the load job in the function named createMiniLoad.
But sometimes, the load job hasn't been created in catch statement. It will throw the NullPointerException when the load job is cancelled.
This commit fix this bug.
* Fix bug that <=> operator and in operator get wrong result
* Add some comment to get_result_for_null
* Add an new Binary Operator to replace is_safe_for_null for handleing '<=>' operator
* Add EQ_FOR_NULL to TExprOpcode
* Remove macro definition last backslash
Use same UUID as query ID and load ID of a load execution plan.
Each load execution plan has a load ID, and as a plan, there is also a query ID.
We can use same UUID as query ID and load ID, for tracing the load process more easily.
Change the load ID when retrying a load execution plan.
When a load execution plan retry, the load ID should be changed, otherwise BE can not
distinguish the old and new load requests.
Cancel the running loading task when cancelling the broker load.
When user cancel a broker load, the running loading task should also be cancelled, or
it may occupies the worker thread for a long time.
Remove the unnecessary query report when doing load execution plan.
Only the last query report is needed.
Add a new BE config tablet_writer_rpc_timeout_sec.
It is used for RPC of tablet sink. The default is 600 seconds. which is long enough for flushing
about 6GB data. The long timeout config will reduce the possibility of encountering fail to send batch error when loading.
Use streaming_load_max_mb instead of mini_load_max_mb in BE config.
Add more logs for tracing a broker load process easily.
The default value of null is forbidden in insert into stmt while null column has not been mentioned in stmt.
This is a bug because the unmentioned column has default value. The values should be inserted successfully although the default value is null.
So the column may simply be not assigned default value when the column is not allowed null and the default value of column is null.
Add dict encoding page for binary/string type data.
Construct a dict for original data, and save encoded id instead of
origin data to save space. If the dict is too big, then will automatically fall
back to plain encoding.
The catalogue of load docs:
---- load-manual.md
---- broker-load-manual.md
---- insert-into-manual.md
---- stream-load-manual.md
This commit also changes max/min_stream_load_timeout to max/min_load_timeout.
The old config named stream_load_timeout means the max timeout suited for all types of load.
So the config name has been changed.
The txn attachment maybe null when broker load has been cancelled without attachment.
The end log of broker load has been record but the callback id of txnState hasn't been removed
So the callback of txn is executed when log of txn aborted is replayed.
The reason for validate failure is the cloned file's names
may conflict and load segment read file througth cache and
cache key is file name, so index may read wrong file. The
solution is load index without use file handle cache.
The Operator wants to known when the job being scheduled as PENDING
and LOADING. And how long it takes to finish these sub states.
Also add 2 metrics on BE to monitor the memtable's flush time.
`memtable_flush_total` and `memtable_flush_duration_us`
Use RowwiseIterator to uniform all data fetch in storage engine.
All objects in storage engine can be read in iterator format.
For example: Segment, Rowset.
This patch implement two generic iterators: UnionRowwiseIterator,
MergeRowwiseIterator. These two class will add iterator as its inputs.
To implement iterators, we define a new class RowBlockV2, all data read
from iterator is in this format. We define a new class other than use
old version's RowBlock is because we want to keep old code work
normally.
When doing rollup, seek_columns equals to the complete set of tablet's columns.
There is no necessity to set it.
Related to commit 36df6ebe4e5f0abd3f07c1e454710590f1de23c7
The `_profile` in OlapTableSink may not be initialized if `prepare()`
method is not called. So when close the OlapTableSink, we should
check if `_profile` is initialized.
When query result reach limit, the Coordinator in FE will send a cancel
request to BE to cancel the query. And when being cancelled, BE will report
query status to FE for debug purpose. But actually it is not necessary
and will generate too many logs.
So I add a CancelReason to distinguish the difference between 'normally'
cancellation and 'internal error' cancellation. if 'normally' cancelled,
no status will be reported from BE.
When query reach limit, or user cancel it actively, it is being cancelled 'normally'.
Otherwise, the query is cancelled due to internal error, which will need
a report from BE.
The function of miniLoadBegin will return the txn_id.
If the backend sends the duplicated request to frontend, frontend will return the txn_id which was created by the same mini load.
The issue is that frontend returns the txn_id when the last same request hasn't been begun the txn.
The frontend returns the zero which is initialized txn_id and the be could not execute the load plan with a error txn_id.
The commit conbines the `createLoadJob` and `execute` together in the write lock. It protects the atomicity of `create` and `beginTxn`.
So the duplicated request cannot get the txn id before the last same request is finished.
