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 | 353,651,444 | 1,689 | alexey-milovidov | @silviucpp Please don't be afraid.
Let me accept this revert and I will wait for your next PR, where you can add this feature back, with even more clear and understandable code. I appreciate the importance of this feature and I think that it should be merged anyway. | 2017-12-22T18:25:18 | 2017-12-22T18:25:18 | {} | MEMBER |
ClickHouse | ClickHouse | 353,655,618 | 1,676 | alexey-milovidov | > Для "уровня 2" добавил возможность создавать/удалять и писать во временные таблицы. Возможно стоит для этого добавить еще какой-то лимит который будет отвечать за это?
Решение Ок. Возможность работать с временными таблицами в режиме readonly - полностью естественно. Сейчас посмотрю код... | 2017-12-22T18:51:42 | 2017-12-22T18:51:42 | {} | MEMBER |
ClickHouse | ClickHouse | 353,702,129 | 1,510 | alexey-milovidov | This should be fixed in the latest 1.1.54327 release,
but there is still minor chance, that small performance regression exist and we cannot catch it. | 2017-12-23T02:36:34 | 2017-12-23T02:36:34 | {} | MEMBER |
ClickHouse | ClickHouse | 353,727,958 | 1,661 | Slind14 | I have been trying to give the CollapsingMergeTree a run. Unfortunately, as soon as I add "FINAL" it doesn't return anything.
```
CREATE TABLE impressions
ENGINE = CollapsingMergeTree(`date`, (`impressionUUID`, `date`), 8192, sign)
AS SELECT
toInt8(-1) AS `sign`,
toFixedString(UUIDStringToNum(`impressionUUID`),... | 2017-12-23T14:09:49 | 2017-12-23T15:15:14 | {} | NONE |
ClickHouse | ClickHouse | 354,095,888 | 1,708 | ztlpn | Take a look at the [star schema benchmark](https://www.cs.umb.edu/~poneil/StarSchemaB.pdf) which is pretty similar to TPC-H. Here is a [howto guide](https://clickhouse.yandex/docs/en/getting_started/example_datasets/star_schema.html) and [some results](https://www.percona.com/blog/2017/06/22/clickhouse-general-analytic... | 2017-12-27T10:54:37 | 2017-12-27T10:54:37 | {} | CONTRIBUTOR |
ClickHouse | ClickHouse | 353,397,571 | 1,672 | lamberken | @filimonov, hi. I can't get right date, version 1.1.54310
#### select version()
#### create table
```
CREATE TABLE test (
eventdate Date DEFAULT toDate(now()),
bodyid Int32,
views Int32
) ... | 2017-12-21T16:39:45 | 2017-12-21T16:39:45 | {} | CONTRIBUTOR |
ClickHouse | ClickHouse | 353,461,145 | 1,661 | filimonov | > If the update occurs usually within 5 minutes, will merging be fast? E.g. if I run a realtime query without final, will only the last hour be off, since everything older is very likely to be merged by then?
There are no garantee that it will merge the parts as you would expect.
So, lets say your A parts contain... | 2017-12-21T21:23:34 | 2017-12-21T21:45:19 | {} | CONTRIBUTOR |
ClickHouse | ClickHouse | 353,506,617 | 1,672 | lamberken | @alexey-milovidov :+1:
ok | 2017-12-22T02:24:26 | 2017-12-22T02:24:26 | {} | CONTRIBUTOR |
ClickHouse | ClickHouse | 353,506,714 | 1,400 | alexey-milovidov | 7a15be8e6 | 2017-12-22T02:25:15 | 2017-12-22T02:25:15 | {
"+1": 1
} | MEMBER |
ClickHouse | ClickHouse | 353,690,735 | 1,678 | alexey-milovidov | Added aggregate function `sumKahan`. It will sum floating point numbers more accurately,
though it is slower (up to 1.5 times) and it is also non-deterministic.
https://github.com/yandex/ClickHouse/commit/609133ea01803379e340e4c9658d276ea7ac4f9e#diff-c6bed5bbf8de99a2db278556cf47bb18 | 2017-12-22T23:35:38 | 2017-12-22T23:35:38 | {} | MEMBER |
ClickHouse | ClickHouse | 353,692,156 | 1,690 | vavrusa | I patched the IProfilingInputStream to check limits before pulling block from the stream, this should fix the last message missing.
```
CREATE MATERIALIZED VIEW test_kafka_ck_mat
ENGINE = MergeTree(day, (day, key), 8192) AS
SELECT
toDate(toDateTime(timestamp)) AS day,
key,
xxx
FROM test_kafka_ck
... | 2017-12-22T23:54:04 | 2017-12-22T23:54:51 | {} | CONTRIBUTOR |
ClickHouse | ClickHouse | 353,861,927 | 863 | TimonKK | +1 | 2017-12-25T11:14:31 | 2017-12-25T11:14:31 | {} | NONE |
ClickHouse | ClickHouse | 353,994,704 | 1,704 | alexey-milovidov | Thank you! | 2017-12-26T17:52:30 | 2017-12-26T17:52:30 | {} | MEMBER |
ClickHouse | ClickHouse | 354,024,218 | 11 | alexey-milovidov | > Since CH already bundles Clang, how about UDFs in C/CPP? Scalar functions could be a good start...
First we will try to add dynamic code generation for compositions of simple functions.
For example, `a * b + c` will be compiled to single loop.
Only after we will think about adding C++ UDF. | 2017-12-26T23:23:07 | 2017-12-26T23:23:17 | {
"+1": 1
} | MEMBER |
ClickHouse | ClickHouse | 353,420,011 | 1,685 | alexey-milovidov | You can also attach your existing file after the table is created, but it must be:
- located exactly in the path of the table;
- named like `data.format`, where format is, for example, `TSV`;
- accessible for the `clickhouse` user. | 2017-12-21T18:15:28 | 2017-12-21T18:15:49 | {} | MEMBER |
ClickHouse | ClickHouse | 353,421,555 | 1,672 | alexey-milovidov | f11740653 | 2017-12-21T18:22:20 | 2017-12-21T18:22:20 | {} | MEMBER |
ClickHouse | ClickHouse | 353,698,382 | 1,644 | alexey-milovidov | Currently we use fixed index_granularity for some arithmetic when reading data.
Worth to note, that it's better to have variable size (adaptive) index granularity.
For example, it's better to place "marks" not after each n-th row, but after some amount of data in bytes.
If we do it, it will be possible to also h... | 2017-12-23T01:19:56 | 2017-12-23T01:19:56 | {} | MEMBER |
ClickHouse | ClickHouse | 353,701,892 | 1,530 | alexey-milovidov | Short answer: use `GLOBAL IN` instead of `IN`.
Details: https://clickhouse.yandex/docs/en/query_language/queries.html?highlight=subqueries#distributed-subqueries | 2017-12-23T02:30:55 | 2017-12-23T02:30:55 | {} | MEMBER |
ClickHouse | ClickHouse | 353,838,068 | 1,657 | lamberken | @KochetovNicolai @ztlpn hi, how to solve the problem?
and if I create a new container also name 'CH-6', the query can go through | 2017-12-25T06:47:30 | 2017-12-25T07:13:02 | {} | CONTRIBUTOR |
ClickHouse | ClickHouse | 353,878,740 | 1,650 | alexey-milovidov | Now all functions working with DateTime, add a "tag" with timezone to return type.
This is intended for correct formatting of results.
DateTime data types with different "tags" should not be considered as different types.
Fixed in master. | 2017-12-25T16:24:50 | 2017-12-25T16:24:50 | {} | MEMBER |
ClickHouse | ClickHouse | 353,395,910 | 1,684 | igor-sh8 | If I set load_balancing='in_order' does it applies only to SELECT queries or also to INSERT queries ?
If I have <internal_replication>false</internal_replication> will the data be always written to the first replica and automatically replicated to second one ?
| 2017-12-21T16:33:18 | 2017-12-21T16:33:18 | {} | NONE |
ClickHouse | ClickHouse | 353,419,500 | 1,684 | alexey-milovidov | With `internal_replication = false`, the setting `load_balancing` doesn't matter for INSERTs as all replicas are used. | 2017-12-21T18:13:00 | 2017-12-21T18:13:00 | {} | MEMBER |
ClickHouse | ClickHouse | 353,420,883 | 1,672 | alexey-milovidov | Both client and server must be up to date.
Because when you use clickhouse-client, the server send binary data and client formats data by itself.
And older clients cannot format dates past 2038.
```
:) CREATE TABLE test ( eventdate Date DEFAULT toDate(now()), bodyid Int32, views Int32 ) ENGINE = Merge... | 2017-12-21T18:19:21 | 2017-12-21T18:19:21 | {} | MEMBER |
ClickHouse | ClickHouse | 353,426,147 | 1,665 | alexey-milovidov | We may implement a function to format floats with specified rules, and a setting for default float formatting. | 2017-12-21T18:42:45 | 2017-12-21T18:42:52 | {} | MEMBER |
ClickHouse | ClickHouse | 353,455,658 | 1,661 | filimonov | If it's millions and 50% of the records are "seen" dictionary is definitely not an option.
I still think that CollapsingMergeTree / ReplacingMergeTree will be better for your case. It's more universal (with that engines you can also make other updates), and if you will update only fresh data (as you said: second up... | 2017-12-21T20:55:57 | 2017-12-21T21:00:35 | {} | CONTRIBUTOR |
ClickHouse | ClickHouse | 353,701,152 | 1,630 | alexey-milovidov | Travis is non-free if the build takes more than half an hour on single core.
> because optional is still under experimental namespace
Try to install clang with libc++. | 2017-12-23T02:14:59 | 2017-12-23T02:14:59 | {} | MEMBER |
ClickHouse | ClickHouse | 353,877,489 | 1,699 | alexey-milovidov | Great! | 2017-12-25T15:59:48 | 2017-12-25T15:59:48 | {} | MEMBER |
ClickHouse | ClickHouse | 353,917,569 | 1,690 | tobad357 | @vavrusa will try to build an test with the new changes
Re the flush, I had added that to the local test but forgot to update the code in the example I added to the previous issue. I did however verify that the messages were in the topic with kafka cat but not in clickhouse
I will revert once I've worked out how t... | 2017-12-26T04:56:11 | 2017-12-26T04:56:11 | {} | NONE |
ClickHouse | ClickHouse | 353,966,478 | 1,696 | KochetovNicolai | Thank you for detailed explanation!
Found out that if `range_min` or `range_max` is NULL that they can contain trash.
Btw, NULL values in dictionaries sources are being replaced by `<null_value>...</null_value>`
https://github.com/yandex/ClickHouse/pull/1702 | 2017-12-26T13:02:42 | 2017-12-26T13:02:42 | {
"+1": 2
} | MEMBER |
ClickHouse | ClickHouse | 353,997,827 | 1,633 | alexey-milovidov | With your change, it will work poor for INSERT queries.
The user may pass `INSERT INTO table FORMAT TSV ...` and then terabytes of data.
`max_query_size` parameter should limit only `INSERT INTO table FORMAT TSV`,
and INSERT data should be processed in O(1) memory. | 2017-12-26T18:23:33 | 2017-12-26T18:23:33 | {} | MEMBER |
ClickHouse | ClickHouse | 353,358,156 | 1,682 | KochetovNicolai | Also you can use subquery as a workaround:
```SQL
INSERT INTO test_number SELECT * FROM (WITH number * 2 AS num2 SELECT number, num2 FROM system.numbers LIMIT 3)
``` | 2017-12-21T14:02:27 | 2017-12-21T14:02:27 | {
"+1": 2
} | MEMBER |
ClickHouse | ClickHouse | 353,464,157 | 1,661 | filimonov | >Btw. do you know if there is any gain from using NVMe SSDs (4-5 times the iops and seq. read) over SATA SSDs?
Are you asking about CLickhouse or in general? :) In clickhouse disk reads happens only for "cold" data. If you have a lot of memory (>128Gb), and most of parts which are actively used fit in memory - disk ... | 2017-12-21T21:38:51 | 2017-12-21T21:44:33 | {
"+1": 5
} | CONTRIBUTOR |
ClickHouse | ClickHouse | 353,516,955 | 1,688 | alexey-milovidov | 622f0b62e038e1b8fc1d10d3d2e782ce4e553105
b9ca36e5ad530e9488cc501d522f5aad8126d5ed | 2017-12-22T04:00:30 | 2017-12-22T04:00:30 | {} | MEMBER |
ClickHouse | ClickHouse | 353,655,053 | 1,556 | alexey-milovidov | Missing functional test and comments.
I am trying to figure out - why the code is written exactly this way.
> auto out = std::make_shared<PushingToViewsBlockOutputStream>(
database_table.first, database_table.second, dependent_table, *views_context, ast_ptr);
Recently we passed empty ASTPtr object instead of a ... | 2017-12-22T18:48:05 | 2017-12-22T18:48:41 | {} | MEMBER |
ClickHouse | ClickHouse | 353,660,353 | 1,676 | alexey-milovidov | Please send separate pull request with:
- functional test for `toStartOfFifteenMinute`;
- functional test for `CREATE`, `INSERT` and `DROP` into temporary tables without readonly, with readonly = 1 and 2; and with non-temporary tables and readonly = 1. | 2017-12-22T19:22:49 | 2017-12-22T19:22:59 | {} | MEMBER |
ClickHouse | ClickHouse | 353,696,866 | 1,691 | vavrusa | @tobad357 can you rebuild ClickHouse from this PR and run your tests? It'd be great if you could help out and convert your tests into an integration test here https://github.com/yandex/ClickHouse/tree/master/dbms/tests/integration (we'd need a docker compose with Kafka, kafkacat, and configuration) | 2017-12-23T00:54:59 | 2017-12-23T00:54:59 | {} | CONTRIBUTOR |
ClickHouse | ClickHouse | 353,697,806 | 1,645 | alexey-milovidov | Try to set merge_tree settings in config:
```
<!-- Settings to fine tune MergeTree tables. See documentation in source code, in MergeTreeSettings.h -->
<!--
<merge_tree>
<max_suspicious_broken_parts>5</max_suspicious_broken_parts>
</merge_tree>
-->
```
The list of all settings can be ... | 2017-12-23T01:10:26 | 2017-12-23T01:10:26 | {} | MEMBER |
ClickHouse | ClickHouse | 353,702,843 | 1,640 | alexey-milovidov | Поддержку массивов в словарях мы хотели бы сделать,
но парсинг массивов в формате PostgreSQL, возможно, придётся вынести в отдельную задачу,
либо реализовать возможность постпроцессинга значений, загруженных из удалённого источника, с помощью выражений в ClickHouse - тогда можно будет парсить эти массивы при загрузке... | 2017-12-23T02:54:03 | 2017-12-23T02:54:03 | {} | MEMBER |
ClickHouse | ClickHouse | 353,750,990 | 1,689 | ztlpn | Here are some details:
> what are the steps that makes replication failing ?
To reproduce the DROP TABLE bug, just create a single Replicated table and then drop it. During the drop the ReplicatedMergeTreeAlterThread task handle will get deactivated, but then the ZooKeeper watch will fire (because the table metad... | 2017-12-23T22:09:53 | 2017-12-23T22:09:53 | {} | CONTRIBUTOR |
ClickHouse | ClickHouse | 353,883,526 | 1,701 | proller | user request "у меня более сложное дерево и этот хостнейм посередине мне всё портит" | 2017-12-25T18:15:43 | 2017-12-25T18:15:43 | {} | CONTRIBUTOR |
ClickHouse | ClickHouse | 353,945,292 | 1,657 | lamberken | @KochetovNicolai @proller @ztlpn can you take time to solve this question? thank you :smiley:
steps:
1. create CH-6, CH-7, CH-8, CH-9 containers
2. create replica + distribute tables
3. stop CH-6
4. insert data to CH-9's distribute table
throw error --> 'Temporary DNS error while resolving'
5. start CH-6
... | 2017-12-26T09:29:34 | 2017-12-26T10:01:11 | {} | CONTRIBUTOR |
ClickHouse | ClickHouse | 353,427,154 | 1,665 | filimonov | I like using printf / sprintf in any language i code, so if some kind of sprintf can be adopted in clickhouse - it woukd be really cool. | 2017-12-21T18:47:10 | 2017-12-21T18:47:10 | {} | CONTRIBUTOR |
ClickHouse | ClickHouse | 353,503,187 | 1,321 | alexey-milovidov | 0972d7586 | 2017-12-22T01:56:44 | 2017-12-22T01:56:44 | {} | MEMBER |
ClickHouse | ClickHouse | 353,690,575 | 1,690 | vavrusa | So I tried to reproduce your setup and there are a couple things going on:
1. The producer script doesn't produce (or flush) all messages, I'm noticing a few messages missing when I read them with kafkacat
```
$ kafkacat -b localhost:9092 -G test_group_7 test_ck > out; wc -l out
% Group test_group_7 rebalanced ... | 2017-12-22T23:33:46 | 2017-12-22T23:33:46 | {} | CONTRIBUTOR |
ClickHouse | ClickHouse | 353,701,555 | 1,600 | alexey-milovidov | Added in master. | 2017-12-23T02:23:27 | 2017-12-23T02:23:27 | {} | MEMBER |
ClickHouse | ClickHouse | 353,766,167 | 1,692 | alexey-milovidov | Thanks!
I wonder, how did you discover this issue? | 2017-12-24T05:22:28 | 2017-12-24T05:22:28 | {} | MEMBER |
ClickHouse | ClickHouse | 353,441,117 | 1,685 | arturgspb | Все понял, спасибо! | 2017-12-21T19:46:25 | 2017-12-21T19:46:25 | {} | NONE |
ClickHouse | ClickHouse | 353,652,478 | 1,689 | silviucpp | @alexey-milovidov I don't have any problem with this revert :) Important is to have this feature working :)
I will appreciate some feedback on the questions I have in order to be able to start working on this again. And also et me know what else you think that we should change in order to make a better and clear co... | 2017-12-22T18:31:54 | 2017-12-22T18:31:54 | {} | CONTRIBUTOR |
ClickHouse | ClickHouse | 353,661,705 | 1,690 | vavrusa | Can you start clickhouse with debug logging and post logs?
It might lose messages if a rebalance happens and flushing of the last block doesn't finish within the session timeout. | 2017-12-22T19:31:58 | 2017-12-22T19:31:58 | {} | CONTRIBUTOR |
ClickHouse | ClickHouse | 353,702,788 | 1,626 | alexey-milovidov | Fixed in master. Added test. | 2017-12-23T02:52:27 | 2017-12-23T02:52:27 | {} | MEMBER |
ClickHouse | ClickHouse | 353,769,223 | 1,692 | cyfdecyf | I'm trying to figure out whether `SELECT * FROM table LIMIT m, n` will use index granularity to skip data. From server log message and the source code, I guess ClickHouse does not have this optimization even if there's only one part for the MergeTree storage?
The need of `LIMIT m, n` for me is a use case maybe not ... | 2017-12-24T07:09:06 | 2017-12-24T07:09:06 | {} | CONTRIBUTOR |
ClickHouse | ClickHouse | 353,878,465 | 1,693 | alexey-milovidov | That's breaks the documented behaviour when zero date/datetime is formatted with all zeros (and this change obviously breaks tests).
We need to dig deeper: do we need such behaviour, what are pros and cons... | 2017-12-25T16:18:58 | 2017-12-25T16:19:16 | {} | MEMBER |
ClickHouse | ClickHouse | 354,107,964 | 1,633 | silviucpp | `So problem is exactly how you described: "if the query is larger than max_query_size, it is simply cut, and either irrelevant error message will be returned, or even worse, wrong query will be executed". You can test with something like SELECT * FROM table where column1 IN (lot of values here to exceed max_query_size)... | 2017-12-27T12:36:40 | 2017-12-27T12:36:40 | {} | CONTRIBUTOR |
ClickHouse | ClickHouse | 353,652,502 | 1,449 | alexey-milovidov | Still not ready. | 2017-12-22T18:32:04 | 2017-12-22T18:32:04 | {} | MEMBER |
ClickHouse | ClickHouse | 353,659,359 | 1,676 | alexey-milovidov | Missing functional test. | 2017-12-22T19:16:10 | 2017-12-22T19:16:10 | {} | MEMBER |
ClickHouse | ClickHouse | 353,701,484 | 1,616 | alexey-milovidov | The possible way of implementation:
modify `ConfigProcessor` - add interpreting some attribute, like `from_env`, example:
`<interserver_http_host from_env="HTTP_HOST" />` | 2017-12-23T02:22:14 | 2017-12-23T02:22:14 | {} | MEMBER |
ClickHouse | ClickHouse | 353,801,030 | 1,399 | bobrik | @alexey-milovidov, nope, closing. | 2017-12-24T20:03:20 | 2017-12-24T20:03:20 | {} | CONTRIBUTOR |
ClickHouse | ClickHouse | 353,810,015 | 1,544 | alexey-milovidov | Fixed in master. | 2017-12-25T00:40:28 | 2017-12-25T00:40:28 | {
"+1": 1
} | MEMBER |
ClickHouse | ClickHouse | 353,883,250 | 1,701 | alexey-milovidov | Motivation? | 2017-12-25T18:08:49 | 2017-12-25T18:08:49 | {} | MEMBER |
ClickHouse | ClickHouse | 353,884,556 | 1,650 | vavrusa | Thanks @alexey-milovidov ! | 2017-12-25T18:42:16 | 2017-12-25T18:42:16 | {} | CONTRIBUTOR |
ClickHouse | ClickHouse | 353,996,588 | 1,633 | alexey-milovidov | As I understand, the issue is the following: if the query is larger than `max_query_size`, it is simply cut, and either irrelevant error message will be returned, or even worse, wrong query will be executed.
With your changes, `max_query_size` is not checked while reading query.
Wouldn't it be better to read up to ... | 2017-12-26T18:11:37 | 2017-12-26T18:16:46 | {} | MEMBER |
ClickHouse | ClickHouse | 353,425,574 | 1,665 | alexey-milovidov | > May be some adjusting of floats formatting (like in php / perl / go) can make everybody happy.
These are different decisions. As I understand,
in first variant, numbers are formatted in a way, that can be parsed back to the same bit representation,
in second variant, numbers are formatted with precision up to f... | 2017-12-21T18:40:12 | 2017-12-21T18:41:05 | {} | MEMBER |
ClickHouse | ClickHouse | 353,770,233 | 1,399 | alexey-milovidov | Does it reproduce in latest versions? | 2017-12-24T07:41:36 | 2017-12-24T07:41:36 | {} | MEMBER |
ClickHouse | ClickHouse | 353,812,889 | 1,683 | alexey-milovidov | In master version, it is Ok for both queries (returns correct error message):
```
:) INSERT INTO test.test1 SELECT id, name FROM test.test2 ANY INNER JOIN test.test3 USING (id);
INSERT INTO test.test1 SELECT
id,
name
FROM test.test2
ANY INNER JOIN test.test3 USING (id)
Received exception from s... | 2017-12-25T01:34:01 | 2017-12-25T01:34:01 | {
"+1": 1
} | MEMBER |
ClickHouse | ClickHouse | 353,355,594 | 1,684 | KochetovNicolai | Hi.
Yes, you can achieve it. Just create another cluster configuration with only first replica per shard. And create other distributed table on that cluster (only for reading). | 2017-12-21T13:51:35 | 2017-12-21T13:51:35 | {} | MEMBER |
ClickHouse | ClickHouse | 353,540,767 | 1,661 | alex-zaitsev | > Btw. do you know if there is any gain from using NVMe SSDs (4-5 times the iops and seq. read) over SATA SSDs?
It depends on workload. ClickHouse works fine with SATA, especially in RAID. There is definitely the gain when using faster disks. In our experiments, SATA RAID-10 with 10-12 disks is slower than the same ... | 2017-12-22T07:38:00 | 2017-12-22T07:38:00 | {
"+1": 5
} | CONTRIBUTOR |
ClickHouse | ClickHouse | 353,572,242 | 1,684 | igor-sh8 | That's cool. It works for me.
Thanks for help.
| 2017-12-22T10:41:52 | 2017-12-22T10:41:52 | {} | NONE |
ClickHouse | ClickHouse | 353,611,101 | 1,689 | silviucpp | Hello @ztlpn,
I started to work on this because I have a scenario where I store 100k+ tables on a single node (a cloud solution)
And at this level clickhouse it's not working properly. there are over 600k threads only on the replication tables. And this threads are not IDLE most of the time as you might believe. ... | 2017-12-22T14:43:15 | 2017-12-22T14:43:30 | {} | CONTRIBUTOR |
ClickHouse | ClickHouse | 353,652,720 | 1,408 | alexey-milovidov | This is experimental change only for investigation purposes. | 2017-12-22T18:33:27 | 2017-12-22T18:33:27 | {} | MEMBER |
ClickHouse | ClickHouse | 353,701,182 | 1,623 | alexey-milovidov | Fixed in master. | 2017-12-23T02:15:43 | 2017-12-23T02:15:43 | {} | MEMBER |
ClickHouse | ClickHouse | 353,836,613 | 1,657 | lamberken | my question is ck-6 ck-8 are relicas, ck-8 is alive, why still throw error?
why the query request doesn't return to ck-8? | 2017-12-25T06:28:27 | 2017-12-25T06:28:27 | {} | CONTRIBUTOR |
ClickHouse | ClickHouse | 354,024,428 | 74 | alexey-milovidov | ODBC driver is moved to separate repository:
https://github.com/yandex/clickhouse-odbc
There you will find all details about building on Linux, Windows, Mac. | 2017-12-26T23:25:39 | 2017-12-26T23:25:39 | {} | MEMBER |
ClickHouse | ClickHouse | 354,038,006 | 520 | bobrik | Not sure if it's the right issue to complain, but during lengthy recovery all workers are busy replicating and merge queue never decreases.
| 2017-12-27T02:12:36 | 2017-12-27T02:12:36 | {} | CONTRIBUTOR |
ClickHouse | ClickHouse | 353,383,938 | 1,684 | KochetovNicolai | You can set `load_balancing='in_order'` setting and replicas will be accessed in the same order as they are specified in cluster config.
If all replicas in shard are unavailable, exception is thrown. You can enable `skip_unavailable_shards=1` setting in order to skip unavailable data silently. | 2017-12-21T15:46:22 | 2017-12-21T15:46:22 | {} | MEMBER |
ClickHouse | ClickHouse | 353,414,605 | 1,685 | KochetovNicolai | Можем создать таблицу с движком File:
```SQL
:) create table test_file_engie (number UInt64, string String) engine = File(TSVWithNames)
CREATE TABLE test_file_engie
(
number UInt64,
string String
)
ENGINE = File(TSVWithNames)
Ok.
0 rows in set. Elapsed: 0.017 sec.
```
И работать с ней как с... | 2017-12-21T17:51:23 | 2017-12-21T17:51:23 | {} | MEMBER |
ClickHouse | ClickHouse | 353,438,180 | 1,685 | alexey-milovidov | Мотивация для существования таблиц типа File такая:
1. Основная: Они используются в качестве внутреннего механизма для работы программы `clickhouse-local`.
2. Дополнительные возможности: Чтобы мгновенно сделать доступным в сервере небольшой файл в подготовленном формате. Можно использовать для мелких таблиц, если нуж... | 2017-12-21T19:33:43 | 2017-12-21T19:33:43 | {} | MEMBER |
ClickHouse | ClickHouse | 353,457,209 | 1,661 | Slind14 | Thank you very much. Once I have a big dataset with load I'll do benchmarks with different schemas. Unfortunately that isn't the case yet and therefore I'm thankful for your guidance which allows me to get the most out of what I have.
A single table also has the advantage that I can use the join for other things wi... | 2017-12-21T21:03:47 | 2017-12-21T21:03:47 | {} | NONE |
ClickHouse | ClickHouse | 353,779,526 | 1,695 | proller | 1. debian init script already prints this warning: https://github.com/yandex/ClickHouse/blob/master/debian/clickhouse-server.init#L51
2. you can recompile clickhouse on target host without sse4 usage.
or you can use cmake flags:
-DHAVE_SSE41=0 -DHAVE_SSE42=0 -DHAVE_POPCNT=0 DHAVE_PCLMULQDQ_INTRIN=0 -DPCLMULFLAG=' ... | 2017-12-24T11:38:13 | 2017-12-24T11:38:13 | {} | CONTRIBUTOR |
ClickHouse | ClickHouse | 353,837,120 | 1,416 | alexey-milovidov | Fixed in 04eb0b8 | 2017-12-25T06:34:42 | 2017-12-25T06:34:42 | {} | MEMBER |
ClickHouse | ClickHouse | 353,887,353 | 1,697 | proller | Can you show your dictionary config and minimal queries with this dictionary ? | 2017-12-25T19:36:21 | 2017-12-25T19:36:21 | {} | CONTRIBUTOR |
ClickHouse | ClickHouse | 353,934,720 | 1,697 | millin | Dictionary config example:
<dictionaries>
<dictionary>
<name>geolite2_city</name>
<source>
<executable>
<command>tee /usr/share/GeoIP/test.log | /usr/share/GeoIP/get_country.py</command>
<format>JSONEachRow</format>
</executable>
... | 2017-12-26T07:57:26 | 2017-12-26T07:57:26 | {} | CONTRIBUTOR |
ClickHouse | ClickHouse | 353,480,909 | 1,661 | Slind14 | I see, so I'm better off saving on the disks and putting the savings into memory.
Thank you for all the specific information :) | 2017-12-21T23:11:52 | 2017-12-21T23:11:52 | {} | NONE |
ClickHouse | ClickHouse | 353,724,086 | 1,600 | Vladikamira | Thanks a lot! :) | 2017-12-23T12:38:58 | 2017-12-23T12:38:58 | {} | NONE |
ClickHouse | ClickHouse | 353,691,676 | 1,686 | alexey-milovidov | We cannot simply enforce minimum size of merges, because otherwise there is a chance of accumulation of many small parts far below min threshold in total.
For example, you set min threshold to 1 GB and there are 999 parts of 1 MB size each, and they won't be merged.
If I understand correctly, you want that during... | 2017-12-22T23:47:42 | 2017-12-22T23:47:42 | {} | MEMBER |
ClickHouse | ClickHouse | 353,987,861 | 1,693 | zhang2014 | I think I may have misunderstood , For #1599 , More hope that " DROP PARTITION '0000-00-00 ' " is passed ? | 2017-12-26T16:41:54 | 2017-12-26T16:41:54 | {} | CONTRIBUTOR |
ClickHouse | ClickHouse | 353,433,431 | 1,685 | arturgspb | Спасибо, огромное!
А на сколько это быстро работает?
Например можно ли так подложить справочник кампаний/клиентов или статистику по рекламным кампаниям по месяцам? =)
На сколько это будет проигрывать по скорости обычным таблицам и внешним справочникам (да, я знаю, что справочники это key-value)? | 2017-12-21T19:12:49 | 2017-12-21T19:13:06 | {} | NONE |
ClickHouse | ClickHouse | 353,436,737 | 1,685 | alexey-milovidov | Во время SELECT, файл читается и данные парсятся каждый раз из указанного формата.
Данные читаются в один поток. Файл читается целиком - индексов нет.
Таким образом, время обработки любого SELECT не сильно отличается от времени, которое было бы потрачено, если эти данные загрузить в другую таблицу.
Вывод - если ... | 2017-12-21T19:27:13 | 2017-12-21T19:27:13 | {} | MEMBER |
ClickHouse | ClickHouse | 353,509,592 | 1,625 | tobad357 | @GroundTech then I would try just doing a test with dummy data containing 1 column and see if your speed changes. It could be the IO for clickhouse that is slowing things down | 2017-12-22T02:49:51 | 2017-12-22T02:49:51 | {} | NONE |
ClickHouse | ClickHouse | 353,603,958 | 1,625 | tobad357 | @vavrusa I will close this one as this issue is fixed.
However I've noticed occasional dropped messages when running tests across 500k messages | 2017-12-22T14:02:54 | 2017-12-22T14:02:54 | {} | NONE |
ClickHouse | ClickHouse | 354,023,973 | 1,281 | alexey-milovidov | Fixed in master. | 2017-12-26T23:19:17 | 2017-12-26T23:19:17 | {} | MEMBER |
ClickHouse | ClickHouse | 354,024,785 | 1,359 | alexey-milovidov | I have closed some tasks, but it is still far from being clean. | 2017-12-26T23:31:19 | 2017-12-26T23:31:19 | {} | MEMBER |
ClickHouse | ClickHouse | 353,349,614 | 1,677 | KochetovNicolai | Both variants `a =? & b = ?` and `b= ? & a = ?` give the same performance. It's not necessary to sort fields. | 2017-12-21T13:23:26 | 2017-12-21T13:23:26 | {} | MEMBER |
ClickHouse | ClickHouse | 353,361,481 | 1,684 | igor-sh8 | Thanks for advice. This is really an interesting way.
But what about the situation when one of those two nodes falls? As I understand SELECT query returns only a part of data that is stored on one shard. Maybe there are something like weights for the replicas in shard ? | 2017-12-21T14:16:50 | 2017-12-21T14:16:50 | {} | NONE |
ClickHouse | ClickHouse | 353,408,698 | 1,684 | KochetovNicolai | `load_balancing` applies also to INSERT queries. But you can apply setting for single query with SETTINGS keyword.
`internal_replication` is for Distributed table on ReplicatedMergeTree. If `internal_replication=1` distributed table will insert data only into one replica for each shard (according to `load_balancing`... | 2017-12-21T17:25:30 | 2017-12-21T17:25:30 | {} | MEMBER |
ClickHouse | ClickHouse | 353,459,253 | 1,661 | Slind14 | > About best practices and your SELECTs: in clickhouse if you have some typical SELECT pattern - it's better to add columns which are always used in group by to primary key.
So if in your select you're always (or almost always) grouping the data by language and date, - then it will be nice idea to add them at the begi... | 2017-12-21T21:14:06 | 2017-12-21T21:18:04 | {} | NONE |
ClickHouse | ClickHouse | 353,692,307 | 1,656 | alexey-milovidov | "Ordered set" is created only when the set is used for filtering by primary key.
In other cases, there is no need for ordered set (and it will be too expensive in terms of memory usage).
But there is another way, how to read from Set without using its internal data structure...
StorageSet and StorageJoin create ... | 2017-12-22T23:55:38 | 2017-12-22T23:55:38 | {} | MEMBER |
ClickHouse | ClickHouse | 353,351,823 | 1,677 | silviucpp | Thanks a lot ! | 2017-12-21T13:33:42 | 2017-12-21T13:33:42 | {} | CONTRIBUTOR |
ClickHouse | ClickHouse | 353,780,144 | 1,695 | patsevanton | Thank for answer.
I use CentOS and systemd for run Clickhouse.
I think bad idea recompile clickhouse on target host without sse4
May be add CPU flag SSE 4.2 in binary ? | 2017-12-24T11:52:30 | 2017-12-24T11:52:30 | {} | CONTRIBUTOR |
ClickHouse | ClickHouse | 353,877,659 | 1,694 | alexey-milovidov | Thank you! | 2017-12-25T16:02:41 | 2017-12-25T16:02:41 | {} | MEMBER |
ClickHouse | ClickHouse | 353,966,540 | 1,502 | KochetovNicolai | https://github.com/yandex/ClickHouse/pull/1702 | 2017-12-26T13:03:21 | 2017-12-26T13:03:21 | {} | MEMBER |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.