Currently, for merge-on-write unique table, the delete bitmap of a rowset will be calculated during flush phase, commit phase and publish phase. In this PR, we add a special mark in every rowset considered when we calculate delete bitmap in these three phases. Before we finally merge the delete bitmap to the table meta's delete bitmap, we will check if all the rowsets contain the special mark to check if we have considered all the rowsets during the above three phases.
Because the executor can not fail in publish phase if the coordinator have received successful commits info from all the executors, we just print logs if this correctness check failed rather than report a failure.
If there is a core dump here, it may cover up the real stack, if stack trace indicates heap corruption
(which led to invalid jemalloc metadata), like double free or use-after-free in the application.
Try sanitizers such as ASAN, or build jemalloc with --enable-debug to investigate further.
In some cases, the high load of HDFS may lead to a long time to read the data on HDFS,
thereby slowing down the overall query efficiency. HDFS Client provides Hedged Read.
This function can start another read thread to read the same data when a read request
exceeds a certain threshold and is not returned, and whichever is returned first will use the result.
eg:
create catalog regression properties (
'type'='hms',
'hive.metastore.uris' = 'thrift://172.21.16.47:7004',
'dfs.client.hedged.read.threadpool.size' = '128',
'dfs.client.hedged.read.threshold.millis' = "500"
);
be_custom.conf persistence path is ${doris_home}/conf/be_custom.conf, but if we set ${custom_config_dir} is a different path, will cause be can't read be_custom.conf from ${custom_config_dir}.
set be_custom.conf persist path to ${custom_config_dir}.
change some parameter's default value:
brpc_num_threads from -1 to 256
compaction_task_num_per_disk from 2 to 4
compaction_task_num_per_fast_disk from 4 to 8
fragment_pool_thread_num_max from 512 to 2048
fragment_pool_queue_size from 2048 to 4096
---------
Co-authored-by: yiguolei <yiguolei@gmail.com>
Default, if less than or equal 32 core, the following are 128, 128, 10240, 10240 in turn.
if greater than 32 core, the following are core num * 4, core num * 4, core num * 320, core num * 320 in turn
brpc_heavy_work_pool_threads
brpc_light_work_pool_threads
brpc_heavy_work_pool_max_queue_size
brpc_light_work_pool_max_queue_size
configs
1. Because vertical compaction is enabled by default, it consumes less
memory, we can enlarge default value of compaction related configs.
2. Enlarge default value of shard size related to lock.
configs
Bdbje elect timeout is 30 seconds, so we enlarge thrift_rpc_timeout_ms
and txn_commit_rpc_timeout_ms to 60s.
BTW: enlarge bdbje_lock_timeout_second from 1 to 5.
Use weak_ptr to cache the file handle of file segment. The max cached number of file handles can be configured by `file_cache_max_file_reader_cache_size`, default `1000000`.
Users can inspect the number of cached file handles by request BE metrics: `http://be_host:be_webserver_port/metrics`:
```
# TYPE doris_be_file_cache_segment_reader_cache_size gauge
doris_be_file_cache_segment_reader_cache_size{path="/mnt/datadisk1/gaoxin/file_cache"} 2500
```
Introduce libunwind get stack trace, cost is negligible and has line numbers.
use StackTraceCache, PHDRCache speed up, is customizable and has some optimizations.
Other stack trace tools remain: glog, boost, glibc, in case for need.
TODO:
currently support linux __x86_64__, __arm__, __powerpc__, not supported __FreeBSD__, APPLE
Note: __arm__, __powerpc__ not been verified
Support signal handle
libunwid support unw_backtrace for jemalloc
Use of undefined compile option USE_MUSL for later
Add a new BE config `kerberos_ticket_lifetime_seconds`, default is 86400.
Better set it same as the value of `ticket_lifetime` in `krb5.conf`
If a HDFS fs handle in cache is live longer than HALF of this time, it will be set as invalid and recreated.
And the kerberos ticket will be renewed.
* [Improve](dynamic schema) support filtering invalid data
1. Support dynamic schema to filter illegal data.
2. Expand the regular expression for ColumnName to support more column names.
3. Be compatible with PropertyAnalyzer and support legacy tables.
4. Default disable parse multi dimenssion array, since some bug unresolved
