owner stringclasses 14
values | repo stringclasses 14
values | id int64 116k 4.61B | issue_number int32 1 181k | author stringlengths 1 39 | body stringlengths 1 262k | created_at timestamp[us]date 2000-06-06 02:40:44 2026-06-02 20:16:18 | updated_at timestamp[us]date 2000-06-06 02:40:44 2026-06-02 20:22:56 | reactions unknown | author_association stringclasses 5
values |
|---|---|---|---|---|---|---|---|---|---|
ClickHouse | ClickHouse | 531,883,717 | 6,941 | dimarub2000 | Ok, `Expected: Date. Got: UInt64` is a common thing. Branch is needed.
This happens when `from_type_hint` is not provided to `convertFieldToTypeImpl`.
UPD: Will make this branch more understandable tomorrow. | 2019-09-16T17:47:16 | 2019-09-16T18:28:52 | {} | CONTRIBUTOR |
ClickHouse | ClickHouse | 532,033,380 | 6,954 | den-crane | ```
DESCRIBE TABLE test_area_geo
┌─name────┬─type──────────┬─default_type─┬─default_expression─┬─comment─┬─codec_expression─┐
│ area_id │ Int32 │ │ │ │ │
│ area.1 │ Array(UInt64) │ │ │ │ │
│... | 2019-09-17T02:38:25 | 2019-09-17T02:38:25 | {} | CONTRIBUTOR |
ClickHouse | ClickHouse | 532,121,863 | 6,943 | zhang2014 | In https://github.com/yandex/ClickHouse/pull/6814, column stream has been modified to lazy create. If destroyed the column stream after a column was written, I think this might fix this problem? | 2019-09-17T08:38:28 | 2019-09-17T08:39:02 | {} | CONTRIBUTOR |
ClickHouse | ClickHouse | 532,485,095 | 6,966 | amosbird | I cannot reproduce the linking error in https://github.com/yandex/ClickHouse/pull/6878#discussion_r324606073 using `yandex/clickhouse-deb-builder` | 2019-09-18T01:57:50 | 2019-09-18T01:57:50 | {} | COLLABORATOR |
ClickHouse | ClickHouse | 532,566,189 | 6,867 | nicelulu | @akuzm You can try this pr https://github.com/yandex/ClickHouse/pull/6946 test case , it doesn't work well.
```
SELECT count(*)
FROM
(
SELECT groupArray(ids)
FROM test_groupArray
WHERE name = 'test'
GROUP BY name
)
→ Progress: 1.00 rows, 134.22 MB (4.83 rows/s., 648.22 MB/s.) Received except... | 2019-09-18T07:53:25 | 2019-09-18T07:53:25 | {} | CONTRIBUTOR |
ClickHouse | ClickHouse | 532,987,564 | 6,972 | qMBQx8GH | Hi 4ertus2!
Thanks for the replay))
1. select was rewrited in this way (it looks correct for me):
```
SELECT
DB,
argMax(date_start, test_id) AS date_start,
max(test_id) AS last_test,
argMax(event_action, test_id) AS event_action
FROM
(
SELECT
DB,
test_id,
... | 2019-09-19T06:31:06 | 2019-09-19T06:31:06 | {} | CONTRIBUTOR |
ClickHouse | ClickHouse | 533,083,423 | 6,168 | nikitamikhaylov | Thank you! I will fix it. Probably you do some stuff with boost library without synchronization. | 2019-09-19T11:11:53 | 2019-09-19T11:11:53 | {} | MEMBER |
ClickHouse | ClickHouse | 533,095,379 | 6,993 | flyfoxCI | the data directory like ignored_20190709_1820_1835_3_4416 | 2019-09-19T11:52:01 | 2019-09-19T11:52:01 | {} | NONE |
ClickHouse | ClickHouse | 532,316,011 | 6,867 | akuzm | The test case works fine for me:
```
SELECT bitmapCardinality(bitmapAnd(b[1], b[2]))
FROM
(
SELECT groupArray(bitmap) AS b
FROM test.test_bitmap
WHERE (dimension = 'test1') OR (dimension = 'test2')
)
┌─bitmapCardinality(bitmapAnd(arrayElement(b, 1), arrayElement(b, 2)))─┐
│ ... | 2019-09-17T17:17:11 | 2019-09-17T17:17:11 | {} | CONTRIBUTOR |
ClickHouse | ClickHouse | 532,624,403 | 6,867 | akuzm | > @akuzm You can try this pr #6946 test case , it doesn't work well.
Yes, I can reproduce this. Could you test this fix: https://github.com/yandex/ClickHouse/pull/6951 , does it help? For me, it reduces memory usage, and the query requires less than 2 GB to complete. | 2019-09-18T10:28:59 | 2019-09-18T10:28:59 | {} | CONTRIBUTOR |
ClickHouse | ClickHouse | 532,732,314 | 6,972 | KochetovNicolai | I think that mat. views with joins are not properly working in ClickHouse at all. | 2019-09-18T15:16:24 | 2019-09-18T15:16:24 | {} | MEMBER |
ClickHouse | ClickHouse | 532,740,221 | 6,972 | 4ertus2 | Let's investigate a little bit more. Could you ```set enable_debug_queries=1``` and run ```analyze <your_query>```. It should print rewrited query as you did with subselects. It's interesting what would be the difference with your variant. Most common issue is there you have duplicate columns in both JOIN sides and the... | 2019-09-18T15:33:31 | 2019-09-18T15:33:31 | {} | CONTRIBUTOR |
ClickHouse | ClickHouse | 532,843,757 | 6,950 | filimonov | https://clickhouse-test-reports.s3.yandex.net/6950/f35de2aafb65b8d4e5fc3d529635ed36da18c137/pvs_studio_report/pvs-studio-html-report/sources/BlocksBlockInputStream.h_14.html#ln26 relevant? | 2019-09-18T20:02:59 | 2019-09-18T20:02:59 | {} | CONTRIBUTOR |
ClickHouse | ClickHouse | 532,847,126 | 6,971 | alexey-milovidov | It's not ready. | 2019-09-18T20:12:41 | 2019-09-18T20:12:41 | {} | MEMBER |
ClickHouse | ClickHouse | 533,053,485 | 5,813 | filimonov | Side effect - that can lead to using the column which is smaller but which is much more expensive for decompressing.
```sql
CREATE TABLE cnt_test
(
`n` UInt64 CODEC(DoubleDelta, ZSTD(20)),
`n2` UInt64
)
ENGINE = MergeTree()
ORDER BY tuple();
insert into cnt_test SELECT number - 1000 + rand() % 2000 ... | 2019-09-19T09:42:40 | 2019-09-19T09:47:22 | {} | CONTRIBUTOR |
ClickHouse | ClickHouse | 531,806,641 | 6,894 | akuzm | Oops, I missed this.
So, the idea is that our HashMap uses too much memory because of its exponential resize, so we use a sparse hash map that doesn't allocate as much empty memory?
Does our TwoLevelHashMap also have better memory usage?
It would be good to have a reference dictionary and queries, so that we ca... | 2019-09-16T14:37:38 | 2019-09-16T14:37:38 | {} | CONTRIBUTOR |
ClickHouse | ClickHouse | 531,881,057 | 6,914 | maqroll | @nikitamikhaylov I got here (no experience in C ++). Maybe you can use it, it depends entirely on you. Thanks. | 2019-09-16T17:40:07 | 2019-09-16T17:40:07 | {} | CONTRIBUTOR |
ClickHouse | ClickHouse | 532,092,414 | 6,949 | PlatonB | Интересное решение. Но при работе через клиентские библиотеки настройки удобно менять с помощью SET. | 2019-09-17T07:11:52 | 2019-09-17T07:11:52 | {} | NONE |
ClickHouse | ClickHouse | 532,270,921 | 6,459 | nicelulu | > so how can I build a early version?
> where can I get early version original code?
> Thanks.
You can get this branch ```remotes/yandex/19.8``` | 2019-09-17T15:23:12 | 2019-09-17T15:23:12 | {} | CONTRIBUTOR |
ClickHouse | ClickHouse | 532,317,092 | 6,867 | akuzm | The case from the other PR works as well:
```
:) SELECT ignore(groupArray(number)) FROM (SELECT number FROM system.numbers LIMIT 73554433);
┌─ignore(groupArray(number))─┐
│ 0 │
└────────────────────────────┘
1 rows in set. Elapsed: 2.404 sec. Processed 73.60 million rows, 588.78 MB ... | 2019-09-17T17:19:55 | 2019-09-17T17:19:55 | {} | CONTRIBUTOR |
ClickHouse | ClickHouse | 532,407,595 | 6,966 | azat | Cc: @amosbird | 2019-09-17T21:22:56 | 2019-09-17T21:22:56 | {} | MEMBER |
ClickHouse | ClickHouse | 532,591,062 | 6,914 | nikitamikhaylov | Ok, I see. But we need all test to be green. I will do it in some time. | 2019-09-18T08:58:28 | 2019-09-18T08:58:28 | {} | MEMBER |
ClickHouse | ClickHouse | 532,684,137 | 6,894 | azat | [Added `robin_hood::unordered_map`](https://github.com/yandex/ClickHouse/pull/6894#issuecomment-532570298) | 2019-09-18T13:28:36 | 2019-09-18T13:28:36 | {} | MEMBER |
ClickHouse | ClickHouse | 532,813,452 | 6,948 | filimonov | There are a lot of cases then it's not a type issue: toDate, toString, toStartOfDay, toHour etc.
Typical usecase - timestamp (usual unixtime) of event and separate client timezone field. And we need to get hours when users are the most active (in their tz).
Sidenote: actually storing TZ in string form in that cas... | 2019-09-18T18:39:32 | 2019-09-18T19:58:52 | {} | CONTRIBUTOR |
ClickHouse | ClickHouse | 533,065,082 | 6,880 | yingfeng | Update:
We've implemented bitmap aggregate functions such that some bitmapAnd and bitmapOr could be replaced with aggregate functions. PR will be submitted later, just like the only existing one:`groupBitmapMerge`. The performance is improved a lot, around tens of times faster for huge bitmaps.
For example:
```
... | 2019-09-19T10:14:57 | 2019-09-19T10:14:57 | {} | NONE |
ClickHouse | ClickHouse | 531,919,240 | 6,947 | alexey-milovidov | > With gcc we got huge performance degradation, stack top is always something like
Ok. So, this PR is "approved" to be closed. | 2019-09-16T19:18:30 | 2019-09-16T19:18:43 | {} | MEMBER |
ClickHouse | ClickHouse | 532,220,926 | 6,526 | SaltTan | Another example:
```
CREATE TABLE x_merge as system.settings ENGINE = Merge(system, '^x');
CREATE TABLE x_memory as x_merge ENGINE = Memory;
show create table x_memory
CREATE TABLE x_memory
(
`name` String,
`value` String,
`changed` UInt8,
`description` String,
`_table` ... | 2019-09-17T13:29:07 | 2019-09-17T13:29:07 | {} | CONTRIBUTOR |
ClickHouse | ClickHouse | 532,554,192 | 6,867 | nicelulu | > What is the problem we are fixing?
This case is because a pr https://github.com/yandex/ClickHouse/pull/6908 modified the serialization method of bitmapData a few days ago. | 2019-09-18T07:17:22 | 2019-09-18T07:22:02 | {} | CONTRIBUTOR |
ClickHouse | ClickHouse | 532,816,009 | 6,982 | 4ertus2 | No need to backport (bugfix in new fucntionality - it's not released yet) | 2019-09-18T18:46:15 | 2019-09-18T18:46:15 | {} | CONTRIBUTOR |
ClickHouse | ClickHouse | 533,072,661 | 6,337 | 4ertus2 | How to implement:
1. Add less and greater logic here https://github.com/yandex/ClickHouse/blob/master/dbms/src/Interpreters/CollectJoinOnKeysVisitor.cpp#L72
2. Save kind of comparison here https://github.com/yandex/ClickHouse/blob/master/dbms/src/Interpreters/CollectJoinOnKeysVisitor.cpp#L81
3. Passthrough the kind ... | 2019-09-19T10:39:34 | 2019-09-19T10:40:10 | {} | CONTRIBUTOR |
ClickHouse | ClickHouse | 531,817,072 | 6,949 | PlatonB | Ну тогда пусть в качестве фичреквеста будет инструкция `SET path = any_path`. Пригодится для лёгкого создания и изменения портативных баз данных. | 2019-09-16T15:00:49 | 2019-09-16T15:00:49 | {} | NONE |
ClickHouse | ClickHouse | 531,822,957 | 6,938 | yuzhichang | @KochetovNicolai I confirm that #6911 fixes this issue. Will this fix be merged to 19.14 branch? | 2019-09-16T15:14:06 | 2019-09-16T15:14:06 | {} | CONTRIBUTOR |
ClickHouse | ClickHouse | 532,127,032 | 6,943 | alexey-milovidov | @zhang2014 It should work (if you know that the stream was ended). | 2019-09-17T08:52:32 | 2019-09-17T08:52:32 | {} | MEMBER |
ClickHouse | ClickHouse | 532,609,264 | 6,965 | alexey-milovidov | @akazz, tech debt. | 2019-09-18T09:45:27 | 2019-09-18T09:45:27 | {} | MEMBER |
ClickHouse | ClickHouse | 532,990,460 | 6,972 | qMBQx8GH | I'm attaching server logs of the qieries execution. They could be useful in some way.
1."wrong" results
[BI_ab_test_db2.log](https://github.com/yandex/ClickHouse/files/3629759/BI_ab_test_db2.log)
2. "right" results
[BI_ab_test_db3.log](https://github.com/yandex/ClickHouse/files/3629761/BI_ab_test_db3.log)
| 2019-09-19T06:41:15 | 2019-09-19T06:41:15 | {} | CONTRIBUTOR |
ClickHouse | ClickHouse | 532,106,165 | 5,454 | filimonov | Closing in favour for #6005 | 2019-09-17T07:53:37 | 2019-09-17T07:53:37 | {} | CONTRIBUTOR |
ClickHouse | ClickHouse | 532,188,282 | 6,937 | akuzm | The thread sanitizer CI failure looks unrelated. The performance test didn't finish, but it shouldn't be influenced because we didn't change much. Will merge as is. | 2019-09-17T11:57:05 | 2019-09-17T11:57:05 | {} | CONTRIBUTOR |
ClickHouse | ClickHouse | 532,610,156 | 6,894 | alexey-milovidov | > Maybe google::sparse_hash_map for now (since it is well known and already used by ch), and switch to tsl::sparse_map later?
Ok.
PS. How did you test insert and find speed? What is the benchmark? | 2019-09-18T09:48:03 | 2019-09-18T09:48:03 | {} | MEMBER |
ClickHouse | ClickHouse | 531,944,447 | 6,894 | alexey-milovidov | @akuzm
Our hash table grows 2x and have max fill factor of 50%. The memory overhead is 2 to 4 times.
Sparsehash memory overhead is very close to nothing and it's clearly better.
Other options is to use robin-hood probing that allows to keep max fill factor about 90%*. But it will still have 2x overhead. Also loo... | 2019-09-16T20:27:32 | 2019-09-16T20:28:20 | {} | MEMBER |
ClickHouse | ClickHouse | 532,622,939 | 6,894 | silviucpp | @azat have you tested also https://github.com/martinus/robin-hood-hashing it claims to perform better than more other : https://martin.ankerl.com/2019/04/01/hashmap-benchmarks-01-overview/
| 2019-09-18T10:24:26 | 2019-09-18T10:24:26 | {} | CONTRIBUTOR |
ClickHouse | ClickHouse | 532,763,349 | 6,978 | filimonov | Similar #5473 (but was happening in other circumstances) | 2019-09-18T16:30:44 | 2019-09-18T16:30:44 | {} | CONTRIBUTOR |
ClickHouse | ClickHouse | 533,080,263 | 6,988 | alexey-milovidov | @godfreyd Look at the commits. | 2019-09-19T11:01:19 | 2019-09-19T11:01:19 | {} | MEMBER |
ClickHouse | ClickHouse | 533,102,631 | 6,950 | abyss7 | > https://clickhouse-test-reports.s3.yandex.net/6950/f35de2aafb65b8d4e5fc3d529635ed36da18c137/pvs_studio_report/pvs-studio-html-report/sources/BlocksBlockInputStream.h_14.html#ln26 relevant?
Yep, that's wrong. Anyway I will remove all unrelated changes including these ones. | 2019-09-19T12:15:53 | 2019-09-19T12:15:53 | {} | CONTRIBUTOR |
ClickHouse | ClickHouse | 532,144,413 | 6,948 | alexey-milovidov | @filimonov It cannot be fixed, because time zone is characteristic of data type, not value.
The value is stored as unix timestamp and independent of time zone. Time zone affects operations on DateTime components and converson to and from string. | 2019-09-17T09:38:14 | 2019-09-17T09:38:14 | {} | MEMBER |
ClickHouse | ClickHouse | 532,260,269 | 6,947 | alesapin | latest commit doesn't affect anything | 2019-09-17T14:59:15 | 2019-09-17T14:59:15 | {} | MEMBER |
ClickHouse | ClickHouse | 532,438,380 | 6,968 | azat | >the user should know about this setting and should enable it.
Indeed, plus it is hard to keep all this settings in mind
>Wouldn't it be better to add IStorage::supportSecondaryIndices method and ignore indices if storage is created AS and indices are not supported?
Agree, will update the patch | 2019-09-17T23:19:54 | 2019-09-17T23:19:54 | {} | MEMBER |
ClickHouse | ClickHouse | 532,530,854 | 6,902 | millin | This bug is present in latest stable release v19.14.3.3.
Hope stable version with this patch will be released soon. | 2019-09-18T05:55:22 | 2019-09-18T05:55:22 | {} | CONTRIBUTOR |
ClickHouse | ClickHouse | 532,739,295 | 6,966 | azat | Bundled build uses static jemalloc library, hence no problems (dbms/libclickhouse_common_io linked with libs/libcommon/libcommon, which in case of static jemalloc has sdallocx in it, otherwise it is just linked with jemalloc) | 2019-09-18T15:31:27 | 2019-09-18T15:31:27 | {} | MEMBER |
ClickHouse | ClickHouse | 532,749,390 | 6,966 | amosbird | So you use bundled build with libjemalloc.so, interesting combination. I think the cmake condition should be tightened. Have you checked if it breaks bundled shared build? (in which the static jemalloc library will be linked twice in this pr) | 2019-09-18T15:55:06 | 2019-09-18T15:55:06 | {} | COLLABORATOR |
ClickHouse | ClickHouse | 533,047,888 | 5,813 | filimonov | Should work in 19.13 and above. | 2019-09-19T09:27:29 | 2019-09-19T09:27:29 | {} | CONTRIBUTOR |
ClickHouse | ClickHouse | 533,072,459 | 5,813 | alexey-milovidov | @amosbird has already addressed this issue in one of PRs. We select based on compressed size and uncompressed size. | 2019-09-19T10:38:54 | 2019-09-19T10:38:54 | {} | MEMBER |
ClickHouse | ClickHouse | 531,859,319 | 6,949 | PlatonB | Пример очень удобных портативных БД-подобных структур:
ftp://ftp.1000genomes.ebi.ac.uk/vol1/ftp/release/20130502/supporting/GRCh38_positions/
По ссылке - набор громоздких таблиц и сгенерированных для каждой из них [специальной программой](http://www.htslib.org/doc/tabix.html) индекс-файлов (*.tbi). Было бы прекрасн... | 2019-09-16T16:43:10 | 2019-09-16T16:49:45 | {} | NONE |
ClickHouse | ClickHouse | 532,486,035 | 6,966 | amosbird | I cannot reproduce the linking error you mentioned in https://github.com/yandex/ClickHouse/pull/6878#discussion_r324606073 with `yandex/clickhouse-deb-builder`. And I don't think libjemalloc.a should not be linked here.
```
docker run -it --rm -v /home/amos/git/chorigin/:/build yandex/clickhouse-deb-builder bash
... | 2019-09-18T02:03:08 | 2019-09-18T02:03:48 | {} | COLLABORATOR |
ClickHouse | ClickHouse | 532,903,451 | 6,948 | alexey-milovidov | > storing TZ in string form in that case is suboptimal
Probably the best way is to provide predefined Enum with stable identifiers for every time zone.
> And currently the best workaround i can imagine - is to store current utc offset instead of timezone
Yes, it is viable solution if you store UTC offset at th... | 2019-09-18T23:21:38 | 2019-09-18T23:21:38 | {} | MEMBER |
ClickHouse | ClickHouse | 533,057,485 | 5,596 | excitoon | Smoke checked on [johannesboyne/gofakes3](https://github.com/johannesboyne/gofakes3):
```
excitoon@bionic:~/gofakes3$ go build cmd/gofakes3/main.go
excitoon@bionic:~/gofakes3$ ./main -backend memory -bucket abc -host :8080
excitoon@bionic:~$ clickhouse local -q "insert into table function s3('http://localhost:808... | 2019-09-19T09:53:44 | 2019-09-19T10:21:31 | {
"+1": 1
} | CONTRIBUTOR |
ClickHouse | ClickHouse | 531,813,213 | 6,894 | akuzm | On a second thought, the memory usage of two-level table should only be different when the hash is not uniform and the occupancy of buckets is significantly different. For an ideal hash, memory usage should be the same. | 2019-09-16T14:52:23 | 2019-09-16T14:52:23 | {} | CONTRIBUTOR |
ClickHouse | ClickHouse | 531,917,079 | 6,944 | alexey-milovidov | @Akazz to check test failure that is unrelated to this PR. | 2019-09-16T19:12:36 | 2019-09-16T19:12:36 | {} | MEMBER |
ClickHouse | ClickHouse | 532,138,350 | 6,948 | filimonov | It is known issue / limitation, but maybe it should be fixed someday. | 2019-09-17T09:22:15 | 2019-09-17T09:22:15 | {} | CONTRIBUTOR |
ClickHouse | ClickHouse | 532,269,674 | 6,959 | nicelulu | Related https://github.com/yandex/ClickHouse/issues/6459.
| 2019-09-17T15:20:20 | 2019-09-17T15:20:20 | {} | CONTRIBUTOR |
ClickHouse | ClickHouse | 532,589,876 | 6,914 | maqroll | @nikitamikhaylov Last change compiles and works on basic tests. | 2019-09-18T08:55:40 | 2019-09-18T08:55:40 | {} | CONTRIBUTOR |
ClickHouse | ClickHouse | 532,668,066 | 6,965 | Akazz | > @Akazz, tech debt.
Noted! | 2019-09-18T12:46:44 | 2019-09-18T12:46:44 | {} | CONTRIBUTOR |
ClickHouse | ClickHouse | 532,818,880 | 6,969 | azat | > Style check — Found 4 errors
Should be fixed in 1f34c44521 | 2019-09-18T18:53:38 | 2019-09-18T18:53:38 | {} | MEMBER |
ClickHouse | ClickHouse | 533,011,622 | 6,969 | azat | >Style check — Found 1 errors
Fixed | 2019-09-19T07:48:53 | 2019-09-19T07:48:53 | {} | MEMBER |
ClickHouse | ClickHouse | 531,793,350 | 6,949 | den-crane | Нет такой возможности.
Только с правкой конфигов https://clickhouse.yandex/docs/ru/operations/server_settings/settings/#server_settings-path
| 2019-09-16T14:07:47 | 2019-09-16T14:07:47 | {} | CONTRIBUTOR |
ClickHouse | ClickHouse | 531,797,296 | 3,490 | lapkofear | We have updated prod CH cluster to version 19.14.3.3 and tested our queries. The problem has been solved. Thanks to everyone who helped =) | 2019-09-16T14:16:22 | 2019-09-16T14:16:22 | {} | CONTRIBUTOR |
ClickHouse | ClickHouse | 532,736,685 | 975 | vsmelov | Using of `runningDifference` is absolutely not safe because it breaks on `max_block_count` block borders.
Is there any way to solve it? Or I should use `arrayDifference(groupArray(...))` instead?
```
select
instrument,
arr_timestamp as timestamp,
arr_gap as gap,
from (
select
instrument,
gro... | 2019-09-18T15:25:47 | 2019-09-18T15:25:47 | {} | NONE |
ClickHouse | ClickHouse | 532,785,898 | 6,966 | azat | > So you use bundled build with libjemalloc.so, interesting combination
I don't, I use UNBUNDLED (because otherwise compilation time is too huge).
*And actually it is not possible to build ch in this way without patching cmake (since jemalloc-cmake/CMakeLists.txt has `add_library(jemalloc STATIC)`.*
So let me ... | 2019-09-18T17:29:43 | 2019-09-18T17:29:43 | {} | MEMBER |
ClickHouse | ClickHouse | 532,570,298 | 6,894 | azat | So here are benchmark results (for `uint64_t` key/value pair and 1e7 keys):
| container | insert | find | maxrss |
| ----------------------------- | ----... | 2019-09-18T08:03:48 | 2019-09-18T18:45:48 | {
"+1": 1
} | MEMBER |
ClickHouse | ClickHouse | 532,841,277 | 6,894 | alexey-milovidov | @azat The test is decent (you use preudorandom number for keys so test cannot compare performance on real dataset, because performance on real datasets is always depend on the compromise between hash function quality and the collision resolution method). BTW, what hash function is used in sparsehash by default? If it's... | 2019-09-18T19:56:23 | 2019-09-18T20:00:18 | {} | MEMBER |
ClickHouse | ClickHouse | 532,115,513 | 6,943 | filimonov | > It looks like buffers for all columns are allocated at once, and 20 000 columns lead to about 40 GB of buffers.
So for single insert we need
1 column = 2 Mb.
100 column = 200 Mb,
500 columns = 1Gb
1000 columns = 2 Gb
Beside @nicelulu proposition (which is great): maybe we can decrease the size of preal... | 2019-09-17T08:20:47 | 2019-09-17T08:20:47 | {} | CONTRIBUTOR |
ClickHouse | ClickHouse | 532,498,620 | 6,168 | YiuRULE | Ok, was able to fix most of the bugs, just got a stress test who didn't pass, but I don't have too much information when I read the CI results :( | 2019-09-18T03:04:26 | 2019-09-18T03:04:26 | {} | CONTRIBUTOR |
ClickHouse | ClickHouse | 532,664,266 | 6,959 | akuzm | Does #6974 help? | 2019-09-18T12:36:15 | 2019-09-18T12:36:15 | {} | CONTRIBUTOR |
ClickHouse | ClickHouse | 533,072,598 | 5,813 | alexey-milovidov | BTW, a763a4a is still not implemented. | 2019-09-19T10:39:21 | 2019-09-19T10:39:21 | {} | MEMBER |
ClickHouse | ClickHouse | 533,081,095 | 6,976 | sfod | I've missed two tests using WITH TOTALS without aggregation: 00937_template_output_format and 00378_json_quote_64bit_integers.
I could fix these two tests by removing WITH TOTALS or adding aggregate functions if the fix itself is fine. | 2019-09-19T11:04:03 | 2019-09-19T11:04:03 | {} | CONTRIBUTOR |
ClickHouse | ClickHouse | 533,094,748 | 6,972 | qMBQx8GH | 1) I enabled debug
```
node02 :) set enable_debug_queries=1
SET enable_debug_queries = 1
Ok.
0 rows in set. Elapsed: 0.001 sec.
```
2) ran analyze on:
```
node02 :) analyze SELECT ... | 2019-09-19T11:49:59 | 2019-09-19T11:49:59 | {} | CONTRIBUTOR |
ClickHouse | ClickHouse | 531,841,887 | 6,949 | filimonov | В произвольной папке без изменения конфигов - не будет никогда, потому что безопасность и в произвольные места файловой системы без явного приглашения нельзя.
С изменениями конфигов - можно сейчас выбрать один главный каталог. Можно перенести из него любую папку в любое место и сделать symlink.
С версии 19.15 ... | 2019-09-16T15:58:42 | 2019-09-16T15:58:42 | {} | CONTRIBUTOR |
ClickHouse | ClickHouse | 532,324,746 | 6,959 | akuzm | broken in https://github.com/yandex/ClickHouse/pull/6667 | 2019-09-17T17:39:08 | 2019-09-17T17:39:08 | {} | CONTRIBUTOR |
ClickHouse | ClickHouse | 532,615,859 | 6,894 | azat | >PS. How did you test insert and find speed? What is the benchmark?
In a nutshell they are looks like:
```c
static uint64_t constexpr KEYS = 1e7;
template <class C>
void insert(C &c)
{
for (uint64_t i = 0; i < KEYS; ++i) {
c.insert(std::make_pair(crc64(i), i));
}
}
template <class C>
... | 2019-09-18T10:03:26 | 2019-09-18T10:03:26 | {} | MEMBER |
ClickHouse | ClickHouse | 532,745,209 | 6,973 | KochetovNicolai | Hi. This information you can find in `system.query_log` table (you need to enable query_log in config and set `log_queries=1`). Then if you know query id, run query like
```
SELECT
ProfileEvents.Names,
ProfileEvents.Values
FROM system.query_log
ARRAY JOIN ProfileEvents
WHERE query_id = '<query_id>' and... | 2019-09-18T15:45:14 | 2019-09-18T15:45:14 | {} | MEMBER |
ClickHouse | ClickHouse | 532,803,417 | 6,894 | data-man | @azat I forgot about [parallel-hashmap](https://github.com/greg7mdp/parallel-hashmap) :smile:
Based on [abseil](https://github.com/abseil/abseil-cpp)'s ideas, but absl is a big library. | 2019-09-18T18:14:39 | 2019-09-18T18:39:14 | {} | NONE |
ClickHouse | ClickHouse | 532,816,250 | 6,894 | azat | >I forgot about parallel-hashmap
[Added](https://github.com/yandex/ClickHouse/pull/6894#issuecomment-532570298)
P.S. parallel-hashmap is not about memory efficiency, so I added sparsepp too | 2019-09-18T18:46:52 | 2019-09-18T18:46:52 | {} | MEMBER |
ClickHouse | ClickHouse | 531,841,873 | 6,938 | KochetovNicolai | @yuzhichang yes, we will backport it soon. | 2019-09-16T15:58:40 | 2019-09-16T15:58:40 | {} | MEMBER |
ClickHouse | ClickHouse | 532,144,668 | 6,948 | alexey-milovidov | As an alternative, we may provide non-constant timezone argument for `toString` function. | 2019-09-17T09:38:58 | 2019-09-17T09:38:58 | {} | MEMBER |
ClickHouse | ClickHouse | 532,255,017 | 6,951 | akuzm | `SELECT count() FROM hits_100m_single WHERE NOT ignore(toDate(toString(EventDate)))`
Regresses 1.27x in performance test, no difference in my bencmark. | 2019-09-17T14:48:10 | 2019-09-17T14:48:10 | {} | CONTRIBUTOR |
ClickHouse | ClickHouse | 532,678,773 | 6,966 | azat | You cannot reproduce it using `yandex/clickhouse-deb-builder`, because it has jemalloc<4, and [hence](https://github.com/yandex/ClickHouse/blob/fe9e110a6a6379794444046cefafec05279f8c57/libs/libcommon/include/common/memory.h#L13) `sdallocx` is not used in `libs/libcommon/include/common/memory.h` | 2019-09-18T13:15:17 | 2019-09-18T13:15:17 | {} | MEMBER |
ClickHouse | ClickHouse | 532,852,523 | 6,968 | azat | >Wouldn't it be better to add IStorage::supportSecondaryIndices method and ignore indices if storage is created AS and indices are not supported?
Hm, we do not have Storage yet during parsing the query (and filling indexes), and during creating Storage we do not have `AS` flag (for now), so I'm thinking about just c... | 2019-09-18T20:28:02 | 2019-09-18T20:28:02 | {} | MEMBER |
ClickHouse | ClickHouse | 532,437,692 | 6,968 | alexey-milovidov | The problem is clear.
The proposed solution with setting has downside: the user should know about this setting and should enable it.
Wouldn't it be better to add `IStorage::supportSecondaryIndices` method and ignore indices if storage is created AS and indices are not supported? | 2019-09-17T23:16:31 | 2019-09-17T23:16:45 | {} | MEMBER |
ClickHouse | ClickHouse | 532,574,061 | 6,965 | azat | >Functional stateless tests (thread) — fail:20, passed:44
this is not me, right? | 2019-09-18T08:14:15 | 2019-09-18T08:14:15 | {} | MEMBER |
ClickHouse | ClickHouse | 532,719,912 | 6,966 | amosbird | @azat but we have jemalloc 5.1.0 in contrib and bundled build links fine | 2019-09-18T14:49:10 | 2019-09-18T14:49:10 | {} | COLLABORATOR |
ClickHouse | ClickHouse | 532,815,476 | 6,966 | amosbird | Yes, I understand what you're doing and why you got that linking error. As long as this pr doesn't break any current CI builds, I'm fine with it. | 2019-09-18T18:44:58 | 2019-09-18T18:44:58 | {} | COLLABORATOR |
ClickHouse | ClickHouse | 533,008,624 | 6,972 | 4ertus2 | Hi Maxim,
The interested queries to compare are “as selects” from create view. Could you run analyze on select part from p.2 (it rewrites joins to subselects), make a view based on the result and check its correctness? | 2019-09-19T07:40:16 | 2019-09-19T07:40:16 | {} | CONTRIBUTOR |
ClickHouse | ClickHouse | 533,098,719 | 6,972 | qMBQx8GH | create view based on analyze result worked after I changed last join line.
from:
`) AS v_b24_hit_event_abcrm_2019 ON (date = date_start) AND (db_name = DB)`
to:
`) AS v_b24_hit_event_abcrm_2019 ON (v_b24_hit_event_abcrm_2019.date = date_start) AND (v_b24_hit_event_abcrm_2019.db_name = DB)`
Then view "produced" c... | 2019-09-19T12:03:11 | 2019-09-19T12:03:11 | {} | CONTRIBUTOR |
ClickHouse | ClickHouse | 531,881,736 | 6,914 | nikitamikhaylov | Ok. | 2019-09-16T17:41:59 | 2019-09-16T17:41:59 | {} | MEMBER |
ClickHouse | ClickHouse | 531,892,268 | 6,949 | proller | Любые настройки можно передать серверу при запуске через -- :
clickhouse-server -- --path=/my/home/dir | 2019-09-16T18:09:27 | 2019-09-16T18:09:27 | {} | CONTRIBUTOR |
ClickHouse | ClickHouse | 532,109,179 | 6,949 | filimonov | Сделайте отдельный маленький сервис который будет делать симлинки по требованию клиента.
Добавление такого фичера в код КХ, мягко говоря, не безопасно. Злоумышленник сможет нанести слишком много "пользы" с помощью этого фичера. | 2019-09-17T08:02:16 | 2019-09-17T08:02:16 | {} | CONTRIBUTOR |
ClickHouse | ClickHouse | 532,128,095 | 6,953 | filimonov | Sounds interesting.
Actually log analysis a VERY popular usecase for ClickHouse.
BUT: usually you have logs to analyze on one hosts (set of hosts), and clickhouse on another.
Because of that there are tools which do something like 'tail' on one host and pass the logs to remote server.
* logstash + lot of '[beat... | 2019-09-17T08:55:13 | 2019-09-17T08:55:13 | {} | CONTRIBUTOR |
ClickHouse | ClickHouse | 532,708,010 | 2,847 | SaltTan | I could not reproduce this on 19.13.3 | 2019-09-18T14:23:11 | 2019-09-18T14:23:11 | {} | CONTRIBUTOR |
ClickHouse | ClickHouse | 532,817,370 | 6,894 | azat | > >It's in question - should we provide the choice of the data structure via attribute or via different layout. I think, different layout is better.
> Agree, since in case of attributes it simply ignores non existing, so you cannot be sure what hashtable will be used, I will update the patch
@alexey-milovidov res... | 2019-09-18T18:49:43 | 2019-09-18T18:49:43 | {} | MEMBER |
ClickHouse | ClickHouse | 532,850,904 | 6,894 | azat | >what hash function is used in sparsehash by default? If it's std::hash
Indeed:
```
$ gg std::hash
sparsehash/dense_hash_map:template <class Key, class T, class HashFcn = std::hash<Key>,
sparsehash/dense_hash_set:template <class Value, class HashFcn = std::hash<Value>,
sparsehash/sparse_hash_map:template <cl... | 2019-09-18T20:23:30 | 2019-09-18T20:23:30 | {} | MEMBER |
ClickHouse | ClickHouse | 502,071,138 | 4,679 | I-akshay | > select * from system.settings where name like 'max_bytes_before_e%';
if i want to change the config how to do that because these configs are not present in config.xml | 2019-06-14T11:19:39 | 2019-06-14T11:19:39 | {} | NONE |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.