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 | 565,147,124 | 8,183 | den-crane | No.
I use
```
cd detached
let i=1;for f in `ls -1` ; do echo $i $f;((i++)); echo "alter table T attach part '$f';"|clickhouse-client ;sleep 1; done
```
In modern CH versions probably a table `system.detached_parts` can be used for automation. | 2019-12-12T19:21:56 | 2019-12-12T19:24:10 | {
"+1": 2
} | CONTRIBUTOR |
ClickHouse | ClickHouse | 565,236,111 | 8,181 | alexey-milovidov | So, maybe we are compressing uninitialized memory? | 2019-12-12T23:39:26 | 2019-12-12T23:39:26 | {} | MEMBER |
ClickHouse | ClickHouse | 565,302,621 | 8,164 | nicelulu | > We can add a test with `SELECT count(1) FROM remote('127.0.0.{1,1,2}')` or something like that?
I added some tests. | 2019-12-13T05:09:19 | 2019-12-13T05:09:19 | {} | CONTRIBUTOR |
ClickHouse | ClickHouse | 565,343,487 | 8,164 | alesapin | and they failed :( | 2019-12-13T08:07:20 | 2019-12-13T08:07:20 | {} | MEMBER |
ClickHouse | ClickHouse | 565,386,686 | 8,159 | gubinjie | I also want to know how to configure the hdfs-engine path. | 2019-12-13T10:22:30 | 2019-12-13T10:22:30 | {} | NONE |
ClickHouse | ClickHouse | 565,431,427 | 8,093 | alexey-milovidov | @huang--yijian
Yes, it is safe. We have removed all the dependencies that may prevent installation.
The `clickhouse` binary itself doesn't have any dependencies except old enough version of `libc` that already exists in the system. | 2019-12-13T12:54:40 | 2019-12-13T12:54:40 | {} | MEMBER |
ClickHouse | ClickHouse | 565,440,051 | 8,199 | Enmk | > Is this code identical to what we already have in master?
almost, I haven't noticed that something like that was already done. However, there are missing specializations in master:
```
if (which.idx == TypeIndex::DateTime64)
// ...
```
Let me fixup the PR to include only that. | 2019-12-13T13:25:11 | 2019-12-13T13:25:11 | {} | CONTRIBUTOR |
ClickHouse | ClickHouse | 565,132,173 | 7,170 | alexey-milovidov | https://github.com/ClickHouse/ClickHouse/commit/575da2357368f283a5d9b450144df5c3733b2943 | 2019-12-12T18:41:58 | 2019-12-12T18:41:58 | {
"laugh": 1
} | MEMBER |
ClickHouse | ClickHouse | 565,211,497 | 8,193 | filimonov | Indeed it is related to the Nested logic.
I think the simpliest option would be just to use some other column naming schema. | 2019-12-12T22:15:49 | 2019-12-12T22:15:49 | {} | CONTRIBUTOR |
ClickHouse | ClickHouse | 565,240,874 | 8,196 | alexey-milovidov | In my debug build, the following query always return 4:
```
$ clickhouse-local --query "SELECT randConstant(CAST(NULL, 'Nullable(Int8)'))"
4
```
That's quite funny :) | 2019-12-12T23:58:14 | 2019-12-12T23:58:14 | {
"laugh": 1
} | MEMBER |
ClickHouse | ClickHouse | 565,241,571 | 8,195 | filimonov | Also https://github.com/ClickHouse/ClickHouse/issues/7906 | 2019-12-13T00:00:46 | 2019-12-13T00:00:46 | {} | CONTRIBUTOR |
ClickHouse | ClickHouse | 565,241,890 | 8,196 | alexey-milovidov | @KochetovNicolai The functions `rand` and `randConstant` are using default implementation for NULLs. But the return type if rand(Nullable(...)) is not Nullable. How's that possible? Please clarify. | 2019-12-13T00:01:57 | 2019-12-13T00:01:57 | {} | MEMBER |
ClickHouse | ClickHouse | 565,398,178 | 8,196 | KochetovNicolai | https://xkcd.com/221/
BTW, for function `rand` it works correct on master.
```
nikkochetovdev.sas.yp-c.yandex.net :) select rand(cast(4 as Nullable(UInt8))) as r, toTypeName(r)
SELECT
rand(CAST(4, 'Nullable(UInt8)')) AS r,
toTypeName(r)
┌─────────r─┬─toTypeName(rand(CAST(4, 'Nullable(UInt8)')))─┐... | 2019-12-13T10:57:17 | 2019-12-13T10:57:17 | {} | MEMBER |
ClickHouse | ClickHouse | 565,323,587 | 8,170 | morigs | Yes, it seems this is duplicate.
I'll leave comment about `as` clause in #6104 and I believe we can close this issue as a duplicate | 2019-12-13T06:49:41 | 2019-12-13T16:17:12 | {} | NONE |
ClickHouse | ClickHouse | 565,429,924 | 8,196 | alexey-milovidov | > Oh, I have a fix for that
I'll close this PR and wait for you. | 2019-12-13T12:49:09 | 2019-12-13T12:49:19 | {} | MEMBER |
ClickHouse | ClickHouse | 565,525,034 | 7,386 | akuzm | _Too slow_ meaning that the functional stateless tests in debug take longer to complete than integration tests, which were the longest before (not including performance tests). | 2019-12-13T17:12:45 | 2019-12-13T17:13:21 | {} | CONTRIBUTOR |
ClickHouse | ClickHouse | 565,176,659 | 5,142 | filimonov | fixed somewhere between 19.6.3 and 19.7.5.29. I will add that test. | 2019-12-12T20:38:09 | 2019-12-12T20:38:09 | {} | CONTRIBUTOR |
ClickHouse | ClickHouse | 565,325,014 | 6,104 | morigs | It also works the same way inside `as` part when using `create table ... as`. See #8170 | 2019-12-13T06:55:03 | 2019-12-13T06:55:03 | {} | NONE |
ClickHouse | ClickHouse | 565,397,268 | 8,159 | alex-krash | Currently HA mode is not supported. | 2019-12-13T10:54:35 | 2019-12-13T10:54:35 | {} | CONTRIBUTOR |
ClickHouse | ClickHouse | 565,418,946 | 7,785 | akuzm | > Can't think of a good way to change current behaviour as ALTER commands can be chained.
Can we forbid the chaining in case there is a FREEZE? Apart from this, are there other reasons to add new syntax? | 2019-12-13T12:08:44 | 2019-12-13T12:08:44 | {} | CONTRIBUTOR |
ClickHouse | ClickHouse | 565,088,504 | 8,179 | akuzm | > I'm slightly afraid of `parst_to_do_names` array size for big tables.
We could do a denormalized table like the undocumented `system.parts_columns`, but that would probably be an overkill. | 2019-12-12T16:47:35 | 2019-12-12T16:47:35 | {} | CONTRIBUTOR |
ClickHouse | ClickHouse | 565,142,332 | 8,184 | alexey-milovidov | FYI you can also use `clickhouse-compressor` tool. | 2019-12-12T19:09:21 | 2019-12-12T19:09:21 | {} | MEMBER |
ClickHouse | ClickHouse | 565,168,552 | 8,174 | filimonov | To find more details you can try to follow back log by the thread id `[ 36 ]` | 2019-12-12T20:16:37 | 2019-12-12T20:16:37 | {} | CONTRIBUTOR |
ClickHouse | ClickHouse | 565,223,904 | 8,191 | olegkv | Thanks again! :) | 2019-12-12T22:56:26 | 2019-12-12T22:56:26 | {} | CONTRIBUTOR |
ClickHouse | ClickHouse | 565,238,073 | 8,185 | alexey-milovidov | hung - known issues, CC @Akazz | 2019-12-12T23:47:00 | 2019-12-12T23:47:00 | {} | MEMBER |
ClickHouse | ClickHouse | 565,245,719 | 8,120 | nvartolomei | @jsding you will have to create a new table and copy data. | 2019-12-13T00:17:17 | 2019-12-13T00:17:17 | {} | CONTRIBUTOR |
ClickHouse | ClickHouse | 565,373,854 | 8,166 | Vxider | I tested the parallel insert with the `hits` dataset, and added the results in the description above. | 2019-12-13T09:44:36 | 2019-12-13T09:44:36 | {} | CONTRIBUTOR |
ClickHouse | ClickHouse | 565,399,890 | 7,790 | bgranvea | ok, I have a test case with ALTER ... DELETE instead of TTL but the result is the same. I add it to the pull request. | 2019-12-13T11:02:13 | 2019-12-13T11:02:13 | {} | CONTRIBUTOR |
ClickHouse | ClickHouse | 565,414,422 | 8,166 | Vxider | Thanks for the review, all have been fixed in the new commit. | 2019-12-13T11:51:57 | 2019-12-13T11:51:57 | {} | CONTRIBUTOR |
ClickHouse | ClickHouse | 565,426,055 | 8,112 | akuzm | query | rps
-- | --
SELECT count() FROM system.numbers WHERE NOT ignore(round(toInt64(number), -2)) | (386154838 rps.)
SELECT count() FROM system.numbers WHERE NOT ignore(bankersRound(toInt64(number), -2)) | (294367764 rps.)
Looks like it's significantly slower for ints.
Overall, I think the code became some... | 2019-12-13T12:34:02 | 2019-12-13T12:34:02 | {} | CONTRIBUTOR |
ClickHouse | ClickHouse | 565,538,683 | 8,205 | alexey-milovidov | Maybe better to allow just `REPLACE DICTIONARY` query?
(atomic dictionary recreate)
Support for all possible properties to ALTER will be too cumbersome. | 2019-12-13T17:48:13 | 2019-12-13T17:48:28 | {
"+1": 2
} | MEMBER |
ClickHouse | ClickHouse | 565,134,100 | 8,120 | den-crane | > The original table is too big and uncontrollable increased, we have to query data daily and combine them in application layer.
You are mistaken. @nvartolomei is right.
You need one MV partitioned by day.
CH SQL partition pruning works almost exactly the same as a query to a separated table.
There is no diff... | 2019-12-12T18:47:32 | 2019-12-12T18:48:20 | {} | CONTRIBUTOR |
ClickHouse | ClickHouse | 565,212,372 | 8,191 | filimonov | https://github.com/ClickHouse/ClickHouse/issues/6287 and https://github.com/ClickHouse/ClickHouse/issues/5131 | 2019-12-12T22:18:46 | 2019-12-12T22:21:50 | {} | CONTRIBUTOR |
ClickHouse | ClickHouse | 565,244,549 | 8,120 | jsding | Thanks for the information.
BTW, is there any way to change the partitioning key? or I have to create a new table, and copy all data into this new table?
Appreciate for your help.
| 2019-12-13T00:12:30 | 2019-12-13T00:12:30 | {} | NONE |
ClickHouse | ClickHouse | 565,490,393 | 8,112 | alexey-milovidov | Current implementation is "round half towards higer absolute value":
```
milovidov-Pro-P30 :) SELECT round(15, -1)
SELECT round(15, -1)
┌─round(15, -1)─┐
│ 20 │
└───────────────┘
1 rows in set. Elapsed: 0.020 sec.
milovidov-Pro-P30 :) SELECT round(-15, -1)
SELECT round(-15, -1)
┌─rou... | 2019-12-13T15:45:59 | 2019-12-13T15:46:17 | {} | MEMBER |
ClickHouse | ClickHouse | 565,059,653 | 8,172 | den-crane | echo "exists a.a"|clickhouse-benchmark
An error occurred while processing query:
exists a.a
DB::Exception: Attempt to read after eof: while receiving packet from localhost:9000 | 2019-12-12T15:39:07 | 2019-12-12T15:41:47 | {
"+1": 1
} | CONTRIBUTOR |
ClickHouse | ClickHouse | 565,538,438 | 8,207 | nikitamikhaylov | Exactly. | 2019-12-13T17:47:29 | 2019-12-13T17:47:29 | {} | MEMBER |
ClickHouse | ClickHouse | 565,539,395 | 8,168 | zhang2014 | That seems to be the current logic. I'm not sure if I missed part of the logic.
https://github.com/ClickHouse/ClickHouse/blob/master/dbms/src/Storages/MergeTree/MergeTreeDataSelectExecutor.cpp#L548
And
https://github.com/ClickHouse/ClickHouse/blob/master/dbms/src/Storages/MergeTree/MergeTreeDataSelectExecutor.... | 2019-12-13T17:50:12 | 2019-12-13T17:50:12 | {} | CONTRIBUTOR |
ClickHouse | ClickHouse | 565,136,513 | 8,169 | 4ertus2 | > So now with the change from t1.c1 as c1 to t1.c1 the outer select fails.
I cannot reproduce. This code works.
```
drop table if exists t1;
drop table if exists t2;
create table t1 (c1 UInt8) Engine = Memory;
create table t2 (c1 UInt8) Engine = Memory;
select c1
from (
select t1.c1
from t1
... | 2019-12-12T18:54:19 | 2019-12-12T18:56:51 | {} | CONTRIBUTOR |
ClickHouse | ClickHouse | 565,209,023 | 8,003 | filimonov | > @filimonov Is this a bug for some particular use case or in general when using kafka engine ?
1) it is already fixed, but fix was not released yet.
2) you hit that bug if your messages in Kafka have no timestamp, or if you use Native format.
> Prevent message duplication when producing Kafka table has any MVs ... | 2019-12-12T22:08:25 | 2019-12-12T22:08:25 | {} | CONTRIBUTOR |
ClickHouse | ClickHouse | 565,241,274 | 8,195 | filimonov | Right now - yes. You can create a distributed table above with special cluster configuration with internal_replication=false to fill 2 (or more) Engine=Join tables simultaniously.
| 2019-12-12T23:59:47 | 2019-12-12T23:59:47 | {} | CONTRIBUTOR |
ClickHouse | ClickHouse | 565,354,391 | 8,198 | filimonov | 1) check table defaults & missing columns.
2) check if datatypes match between Distibuted and underlying table.
3) if aneryting looks correct - please share your table definitions and data sample | 2019-12-13T08:45:57 | 2019-12-13T08:47:57 | {} | CONTRIBUTOR |
ClickHouse | ClickHouse | 565,428,569 | 7,785 | nvartolomei | > Can we forbid the chaining in case there is a FREEZE? Apart from this, are there other reasons to add new syntax?
Breaking change?
We might also output result if there is one alter command and do not output if there are no commands but code might get dirty and behaviour might be a bit surprising to users. | 2019-12-13T12:43:53 | 2019-12-13T12:43:53 | {} | CONTRIBUTOR |
ClickHouse | ClickHouse | 565,430,689 | 8,199 | alexey-milovidov | Is this code identical to what we already have in master? | 2019-12-13T12:51:59 | 2019-12-13T12:51:59 | {} | MEMBER |
ClickHouse | ClickHouse | 565,453,005 | 8,204 | akuzm | So can we have `useDefaultImplementationForNulls` for functions that have non-nullable result type for nullable arguments?
I supported this here: https://github.com/ClickHouse/ClickHouse/pull/7386/files#diff-6afa37fa0db71a6ed14bc0bc87dc771aR128
but maybe I have to add asserts instead. | 2019-12-13T14:06:49 | 2019-12-13T14:06:49 | {} | CONTRIBUTOR |
ClickHouse | ClickHouse | 565,139,599 | 8,184 | den-crane | Not compressed.
clickhouse-client --query="select * from default.TestTable where date='2019-11-11' format Native" | gzip > TestTable.native.gz | 2019-12-12T19:02:11 | 2019-12-12T19:02:28 | {
"+1": 1
} | CONTRIBUTOR |
ClickHouse | ClickHouse | 562,448,477 | 8,059 | excitoon | <s>Based on #7420</s> | 2019-12-06T06:20:08 | 2019-12-12T19:45:05 | {} | CONTRIBUTOR |
ClickHouse | ClickHouse | 565,166,681 | 8,174 | filimonov | Do you try to copy the data from Kafka to usual (non materialized) view? Can you show the structure of your tables? | 2019-12-12T20:11:23 | 2019-12-12T20:11:23 | {} | CONTRIBUTOR |
ClickHouse | ClickHouse | 565,342,112 | 8,093 | huang--yijian | > maybe you can download the clickhouse rpm using `thunder` and transfer the file to your machine and install rpm directly
Is it safe to download and install ClickHouse RPMs by "rpm -Fvh", in order to upgrade ClickHouse to 19.17 or later versions in production ?
It skips dependency check by yum install. | 2019-12-13T08:02:01 | 2019-12-13T08:02:01 | {} | NONE |
ClickHouse | ClickHouse | 565,385,885 | 8,164 | nicelulu | > and they failed :(
I'll fetch the latest code and try it. This test is fine with me locally. | 2019-12-13T10:20:06 | 2019-12-13T10:20:18 | {} | CONTRIBUTOR |
ClickHouse | ClickHouse | 565,138,927 | 8,149 | alexey-milovidov | BTW, we can also quickly compare array offsets for equality as a whole. | 2019-12-12T19:00:36 | 2019-12-12T19:00:36 | {} | MEMBER |
ClickHouse | ClickHouse | 565,193,944 | 8,191 | filimonov | Date vs DateTime.
Try:
```
select count() from test_index where id = '1' and dt >= toDateTime(today())
``` | 2019-12-12T21:26:26 | 2019-12-12T21:26:26 | {} | CONTRIBUTOR |
ClickHouse | ClickHouse | 565,210,180 | 8,170 | filimonov | Sounds familiar: #6104 | 2019-12-12T22:12:01 | 2019-12-12T22:12:01 | {} | CONTRIBUTOR |
ClickHouse | ClickHouse | 565,210,592 | 8,191 | olegkv | toDateTime() solves it, thanks!
Wondering, if implicit conversion from Date to DateTime is done by CH, what does it lose during conversion so that index is not used anymore.. | 2019-12-12T22:13:10 | 2019-12-12T22:13:10 | {} | CONTRIBUTOR |
ClickHouse | ClickHouse | 565,072,317 | 8,169 | vincenzon | Well actually ...
I'm discovering this to be more troubling than I thought. In my constructions I have:
```
select c1
from (
select t1.c1 -- Breaking code was: t1.c1 as c1
from t1
any left join t2 on t1.c1 = t2.c1
)
```
So now with the change from `t1.c1 as c1` to `t1.c1` the outer select fai... | 2019-12-12T16:08:14 | 2019-12-12T16:08:44 | {} | NONE |
ClickHouse | ClickHouse | 565,079,728 | 8,179 | alesapin | I'm slightly afraid of `parst_to_do_names` array size for big tables. | 2019-12-12T16:26:31 | 2019-12-12T16:26:31 | {} | MEMBER |
ClickHouse | ClickHouse | 565,127,874 | 8,169 | vincenzon | Here is a minimal reproducing example:
```
drop table if exists t1;
create table t1 (c1 String) ENGINE = Memory;
insert into t1 values ('A1');
drop table if exists t2;
create table t2 (c1 String, c2 String) ENGINE = Memory;
insert into t2 values ('A2', 'A3');
drop table if exists t3;
create table t3 (c2 ... | 2019-12-12T18:30:03 | 2019-12-12T18:30:03 | {} | NONE |
ClickHouse | ClickHouse | 565,133,679 | 8,182 | akuzm | For the record, this is the last context where I've seen it:
https://clickhouse-test-reports.s3.yandex.net/8173/3630b06cf9ba2b1e82b1a121495402a8e00e25b6/functional_stateless_tests_(memory)/stderr.log
Very simple, just an exception thrown.
```
==199==WARNING: MemorySanitizer: use-of-uninitialized-value
#0... | 2019-12-12T18:46:19 | 2019-12-12T18:46:19 | {} | CONTRIBUTOR |
ClickHouse | ClickHouse | 565,145,005 | 8,169 | vincenzon | Right, that's the problem. I thought your table aliasing trick might subvert it and magically work but no luck. I've added the code to our query generator to check if the join is single or multiple and adapted the generated query to account for the difference. It just seems like a bit of a hack, but it definitely works... | 2019-12-12T19:16:16 | 2019-12-12T19:16:16 | {} | NONE |
ClickHouse | ClickHouse | 565,524,673 | 7,386 | akuzm | The test are too slow now, I'll have to profile them. Will separately commit the fixes I made, without enabling the assert. | 2019-12-13T17:11:49 | 2019-12-13T17:11:49 | {} | CONTRIBUTOR |
ClickHouse | ClickHouse | 565,525,147 | 7,792 | KochetovNicolai | I've added an experimental commit which should simplify code a little bit.
It have passed your new test. If something is wrong, will just revert it. | 2019-12-13T17:13:07 | 2019-12-13T17:13:07 | {} | MEMBER |
ClickHouse | ClickHouse | 565,203,906 | 8,183 | simPod | Perfect! Thanks for the command btw, will use it right now meanwhile. | 2019-12-12T21:53:29 | 2019-12-12T21:53:29 | {} | CONTRIBUTOR |
ClickHouse | ClickHouse | 565,234,248 | 8,181 | alexey-milovidov | It didn't: https://clickhouse-test-reports.s3.yandex.net/8181/51c1a120b86445e5367ad80b65b35bd69af89833/functional_stateless_tests_(memory)/stderr.log | 2019-12-12T23:32:16 | 2019-12-12T23:32:16 | {} | MEMBER |
ClickHouse | ClickHouse | 565,415,035 | 8,180 | filimonov | Sidenote: your pipeline looks suspicious. I'm not sure if `sum_table_mv` will produce the results you expect from it. | 2019-12-13T11:54:14 | 2019-12-13T11:54:14 | {} | CONTRIBUTOR |
ClickHouse | ClickHouse | 565,472,280 | 8,204 | KochetovNicolai | > So can we have useDefaultImplementationForNulls for functions that have non-nullable result type for nullable arguments?
No. It's better to add assert. | 2019-12-13T15:00:16 | 2019-12-13T15:00:16 | {} | MEMBER |
ClickHouse | ClickHouse | 565,535,716 | 7,946 | alexey-milovidov | > but instead of file contents, store S3 URL along with refcount
\+ file size. | 2019-12-13T17:40:03 | 2019-12-13T17:40:25 | {} | MEMBER |
ClickHouse | ClickHouse | 565,190,785 | 8,183 | filimonov | But actually it can be quite useful, especially during migration from MergeTree to Replicated.
Let's change it to feature request. | 2019-12-12T21:17:42 | 2019-12-12T21:17:42 | {} | CONTRIBUTOR |
ClickHouse | ClickHouse | 565,192,951 | 8,175 | filimonov | Related https://github.com/ClickHouse/ClickHouse/issues/5486 | 2019-12-12T21:23:42 | 2019-12-12T21:23:42 | {} | CONTRIBUTOR |
ClickHouse | ClickHouse | 565,233,383 | 8,179 | alexey-milovidov | > I'm slightly afraid of parst_to_do_names array size for big tables.
I'm Ok with that. | 2019-12-12T23:29:21 | 2019-12-12T23:29:21 | {} | MEMBER |
ClickHouse | ClickHouse | 565,402,296 | 8,196 | akuzm | > For rand constant - not
Why? This function looks like it should return some non-null value even for nullable arguments. And in general I can imagine that there exists some function that has a non-nullable result type despite having nullable arguments. | 2019-12-13T11:09:26 | 2019-12-13T11:09:26 | {} | CONTRIBUTOR |
ClickHouse | ClickHouse | 565,537,511 | 8,207 | alexey-milovidov | We just lowered the number of iterations and removed timeout? | 2019-12-13T17:45:02 | 2019-12-13T17:45:02 | {} | MEMBER |
ClickHouse | ClickHouse | 565,129,956 | 8,182 | alexey-milovidov | But we don't use libgcc_s for clang build.
We build our own libunwind and our own libcxxabi. | 2019-12-12T18:35:45 | 2019-12-12T18:35:45 | {} | MEMBER |
ClickHouse | ClickHouse | 565,134,614 | 8,182 | alexey-milovidov | Yes, that's clear that `scan_eh_tab` function is reading data from uninstrumented memory and it's as expected. | 2019-12-12T18:48:57 | 2019-12-12T18:48:57 | {} | MEMBER |
ClickHouse | ClickHouse | 565,183,867 | 7,845 | filimonov | That is not something new. It's duplicate of #3492.The problem appears at any server restart.
Related to #6104 | 2019-12-12T20:58:21 | 2019-12-12T20:58:21 | {} | CONTRIBUTOR |
ClickHouse | ClickHouse | 565,429,824 | 8,196 | alexey-milovidov | > Why? This function looks like it should return some non-null value even for nullable arguments. And in general I can imagine that there exists some function that has a non-nullable result type despite having nullable arguments.
Yes, it should not have "useDefaultImplementationForNulls".
As documented, all argumen... | 2019-12-13T12:48:50 | 2019-12-13T12:48:50 | {} | MEMBER |
ClickHouse | ClickHouse | 565,447,745 | 8,196 | KochetovNicolai | #8204 | 2019-12-13T13:50:31 | 2019-12-13T13:50:31 | {} | MEMBER |
ClickHouse | ClickHouse | 565,466,634 | 8,112 | hczhcz | I would suggest `roundHalfUp`.
And maybe `roundHalfDown`, `roundHalfTowardZero`, and `roundHalfTowardInfinity` if people really need them. | 2019-12-13T14:45:14 | 2019-12-13T14:45:14 | {} | CONTRIBUTOR |
ClickHouse | ClickHouse | 565,111,631 | 8,169 | vincenzon | Well actually actually ...
It looks like the behaviour changes depending on whether the join is a single join or a multiple join.
This:
```
select t1.c1
from t1
left join t2 on t1.c1 = t2.c1
```
returns a column labelled "c1". However,
```
select t1.c1
from t1
left join t2 on t1.c1 = t2.c1
left joi... | 2019-12-12T17:45:43 | 2019-12-12T17:45:43 | {} | NONE |
ClickHouse | ClickHouse | 565,132,912 | 8,182 | alexey-milovidov | I have a hypothesis that this function is parsing exception handling info from mmapped part of our binary and as this mmap happens at startup, it's not instrumented. So, the patch makes sense. | 2019-12-12T18:43:56 | 2019-12-12T18:43:56 | {} | MEMBER |
ClickHouse | ClickHouse | 565,139,374 | 8,169 | vincenzon | Sorry, I'm not a saying the queries don't work. I'm saying the resulting column labels are inconsistent. In the single join case the table prefix is dropped (c1 not t1.c1). In the double join case it is not (t1.c1 not c1). I don't have a preference for either (though if I did I'd expect to be able to enforce it using a... | 2019-12-12T19:01:37 | 2019-12-12T19:01:37 | {} | NONE |
ClickHouse | ClickHouse | 565,143,530 | 8,169 | 4ertus2 | OK. Reproduced. It's not possible to access `t1.c1` via `c1` out of multiple join subquery.
```
select c1
from (
select t1.c1 as c1
from t1
any left join t2 on t1.c1 = t2.c1
any left join t3 on t1.c1 = t3.c1
);
```
If there's no alias it sayes
```
Missing columns: 'c1' while processing que... | 2019-12-12T19:12:32 | 2019-12-12T19:13:32 | {} | CONTRIBUTOR |
ClickHouse | ClickHouse | 565,199,141 | 4,064 | filimonov | It is the duplicate of https://github.com/ClickHouse/ClickHouse/issues/5142
Shoule work starting from 19.7 | 2019-12-12T21:40:22 | 2019-12-12T21:40:22 | {} | CONTRIBUTOR |
ClickHouse | ClickHouse | 565,210,817 | 8,194 | excitoon | This fixes problem introduced in #7670 . | 2019-12-12T22:13:48 | 2019-12-12T22:13:48 | {} | CONTRIBUTOR |
ClickHouse | ClickHouse | 565,234,011 | 8,181 | alexey-milovidov | 00600_replace_running_query - flap. | 2019-12-12T23:31:20 | 2019-12-12T23:31:20 | {} | MEMBER |
ClickHouse | ClickHouse | 565,245,427 | 8,166 | alexey-milovidov | The results are very promising!
It will be also interesting to test on real dataset. What if we will INSERT SELECT `test.hits` dataset (all rows; or with some filter): https://clickhouse.yandex/docs/en/getting_started/example_datasets/metrica/ | 2019-12-13T00:16:08 | 2019-12-13T00:16:08 | {} | MEMBER |
ClickHouse | ClickHouse | 565,246,753 | 8,195 | filimonov | > I just wonder whether the RFC Distributed variant and not the Replicated one?
It's just RFC, very unconcrete for now. But sounds like replicated may be needed too. | 2019-12-13T00:21:24 | 2019-12-13T00:21:44 | {} | CONTRIBUTOR |
ClickHouse | ClickHouse | 565,339,318 | 8,196 | akuzm | Oh, I have a fix for that:
https://github.com/ClickHouse/ClickHouse/pull/7386/files#diff-6afa37fa0db71a6ed14bc0bc87dc771aR129 | 2019-12-13T07:51:28 | 2019-12-13T07:51:28 | {} | CONTRIBUTOR |
ClickHouse | ClickHouse | 565,388,710 | 7,790 | bgranvea | I've reproduced the bug on my test database but still haven't a simple test case.
What happens is that 2 empty parts (0 rows) are merged in background. AggregatingSortedBlockInputStream wrongly generates 1 value for SimpleAggregateFunction columns (see my fix). When the result block is written, there is a check in B... | 2019-12-13T10:28:34 | 2019-12-13T10:28:34 | {} | CONTRIBUTOR |
ClickHouse | ClickHouse | 565,405,152 | 8,162 | akuzm | > Oops, seems like we don't have a gcc 8 build in CI. I'll check if we can add it.
We won't add it now because the CI is overloaded, maybe later if it breaks too often. | 2019-12-13T11:18:49 | 2019-12-13T11:18:49 | {} | CONTRIBUTOR |
ClickHouse | ClickHouse | 565,413,220 | 8,197 | filimonov | Did some manual checks. It looks ok. | 2019-12-13T11:47:21 | 2019-12-13T11:47:21 | {} | CONTRIBUTOR |
ClickHouse | ClickHouse | 565,421,416 | 8,180 | samuelblawrence | Appreciate your responses and look forward to the fix being available.
re. your sidenote - can you elaborate? If I make a normal view then I do see the data I want - e.g.
```
CREATE VIEW v1 as
SELECT gid,
sumState(value) as value
FROM
(
select gid,
argMaxMerge(value) as value
FROM to_table ... | 2019-12-13T12:17:46 | 2019-12-13T12:17:46 | {} | NONE |
ClickHouse | ClickHouse | 565,083,824 | 8,177 | stavrolia | > Test needed.
I forget to make this PR a draft or stay [WIP] in name when decide to add test in another commit. Sorry that I wasted your time on obvious comment. | 2019-12-12T16:36:22 | 2019-12-12T16:36:22 | {} | CONTRIBUTOR |
ClickHouse | ClickHouse | 565,133,230 | 8,169 | 4ertus2 | > returns a column labelled "t1.c1". This seems inconsistent to me and unnecessarily complicated to code around.
It's not good but expected. Aliases are used to implement multiple join logic: first join would be rewritten with aliases and wrapped into subselect. For single query logic you'll get first table unqualif... | 2019-12-12T18:44:56 | 2019-12-12T18:44:56 | {} | CONTRIBUTOR |
ClickHouse | ClickHouse | 565,140,405 | 8,149 | alexey-milovidov | `ColumnArray::hasEqualOffsets` | 2019-12-12T19:04:09 | 2019-12-12T19:04:09 | {} | MEMBER |
ClickHouse | ClickHouse | 565,145,973 | 8,180 | filimonov | Subselects in MV above Kafka table don't work (yet). They will (soon). | 2019-12-12T19:18:57 | 2019-12-12T19:19:50 | {} | CONTRIBUTOR |
ClickHouse | ClickHouse | 563,431,120 | 8,107 | excitoon | <s>Based on #7420</s> | 2019-12-09T20:47:51 | 2019-12-12T19:41:47 | {} | CONTRIBUTOR |
ClickHouse | ClickHouse | 565,242,459 | 8,195 | simPod | Will try, thanks!
The linked RFC is IMO the way to go from here so perhaps can close? I just wonder whether the RFC is only about Distributed variant and not the Replicated one? | 2019-12-13T00:04:10 | 2019-12-13T00:04:10 | {} | CONTRIBUTOR |
ClickHouse | ClickHouse | 565,427,338 | 8,112 | alexey-milovidov | > Overall, I think the code became somewhat complicated. I'd add banker's round as a new mode to RoundingMode, implement each mode for each type, make them accessible from SQL as bankersRound() & roundToInt() or something. And then for backwards compatibility write out our current round() as calls to bankers() for floa... | 2019-12-13T12:39:14 | 2019-12-13T12:40:23 | {} | MEMBER |
ClickHouse | ClickHouse | 565,455,478 | 8,199 | Enmk | > Let me fixup the PR to include only that.
Done | 2019-12-13T14:14:05 | 2019-12-13T14:14:05 | {} | CONTRIBUTOR |
ClickHouse | ClickHouse | 573,157,526 | 8,618 | alexey-milovidov | You don't have to build all targets.
To build clickhouse, just type `ninja clickhouse`. | 2020-01-10T18:43:28 | 2020-01-10T18:43:28 | {
"laugh": 1
} | MEMBER |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.