Repository navigation
update starrocks result to 3.0.0-preview - #72
Conversation
|
@alexey-milovidov please take a look if you are available, thx |
|
I constantly get error on Q32: any ideas why? be.INFO.log.20230213-154110.txt |
|
Hi, @nickitat , very thanks for your feedback. I see that your dictionary name contains ubuntu, it looks like you are testing on an ubuntu system. Currently, StarRocks only supports running on centos. The release package is also compiled on centos, and some strange problems may occur if it runs on other system. We have comments at the beginning of FYI, here is the os version info in my test environment. I have tested many times in this environment and encountered no errors. [ec2-user@ip-172-31-61-73 be]$ cat /etc/os-release
NAME="Amazon Linux"
VERSION="2"
ID="amzn"
ID_LIKE="centos rhel fedora"
VERSION_ID="2"
PRETTY_NAME="Amazon Linux 2"
ANSI_COLOR="0;33"
CPE_NAME="cpe:2.3:o:amazon:amazon_linux:2"
HOME_URL="https://amazonlinux.com/"so,could you verify it on Amazon Linux? In addition, I notice that there are some warning messages in your be log. In order to avoid repeated load, StarRocks has a constraint that the label must be unique for each load, otherwise subsequent load job with the same label will be ignored. I just made an update and added the start time in the label to make it unique, which is convenient for you to perform multiple verifications on the same cluster. please take a look again, thanks:) |
sorry, my bad |
Q28 contains backslashes, specifically the backslash in `\1` is very important for the meaning of the query. However, both backslashes in this query are accidentally removed in almost all benchmark scripts because they used `read` instead of `read -r`. This adds the `-r` flag to all the scripts that didn't have it yet, to make sure that `read` interprets it as a raw `\` character instead of an escape sequence. With this change the REGEXP_REPLACE call doesn't always return the literal string "1", but actually (as intended by the query) the referrer domain. This change generally makes the query significantly slower to execute, because now there are many more groups. I'm not sure how this should be listed in the results UI. It might make sense to ignore the query or something. This problem was found independently in ClickHouse#47 and ClickHouse#72, but fixed in a different way. I'm not sure if the fix that was made for starrocks in ClickHouse#72 is actually correct though, because it seems to fix the issue twice: It replaces `\1` with `\\1`, but it also added the `-r` flag to `read`. Apart from those issues there were already a few benchmarks that were using the `-r` flag: ``` ❯ rg 'while read -r' elasticsearch/run.sh 5:cat 'queries.sql' | while read -r QUERY; do starrocks/run.sh 8:cat queries.sql | while read -r query; do doris/run.sh 8:while read -r query; do crunchy-bridge-for-analytics/run.sh 22: while read -r query; do ```
Q28 contains backslashes, specifically the backslash in `\1` is very important for the meaning of the query. However, both backslashes in this query are accidentally removed in almost all benchmark scripts because they used `read` instead of `read -r`. This adds the `-r` flag to all the scripts that didn't have it yet, to make sure that `read` interprets it as a raw `\` character instead of an escape sequence. With this change the REGEXP_REPLACE call doesn't always return the literal string "1", but actually (as intended by the query) the referrer domain. This change generally makes the query significantly slower to execute, because now there are many more groups. I'm not sure how this should be listed in the results UI. It might make sense to ignore the query or something. This problem was found independently in ClickHouse#47 and ClickHouse#72, but fixed in a different way. I'm not sure if the fix that was made for starrocks in ClickHouse#72 is actually correct though, because it seems to fix the issue twice: It replaces `\1` with `\\1`, but it also added the `-r` flag to `read`. Apart from those issues there were already a few benchmarks that were using the `-r` flag: ``` ❯ rg 'while read -r' elasticsearch/run.sh 5:cat 'queries.sql' | while read -r QUERY; do starrocks/run.sh 8:cat queries.sql | while read -r query; do doris/run.sh 8:while read -r query; do crunchy-bridge-for-analytics/run.sh 22: while read -r query; do ```
Q28 contains backslashes, specifically the backslash in `\1` is very important for the meaning of the query. However, both backslashes in this query are accidentally removed in almost all benchmark scripts because they used `read` instead of `read -r`. This adds the `-r` flag to all the scripts that didn't have it yet, to make sure that `read` interprets it as a raw `\` character instead of an escape sequence. With this change the REGEXP_REPLACE call doesn't always return the literal string "1", but actually (as intended by the query) the referrer domain. This change generally makes the query significantly slower to execute, because now there are many more groups. I'm not sure how this should be listed in the results UI. It might make sense to ignore the query or something. This problem was found independently in ClickHouse#47 and ClickHouse#72, but fixed in a different way. I'm not sure if the fix that was made for starrocks in ClickHouse#72 is actually correct though, because it seems to fix the issue twice: It replaces `\1` with `\\1`, but it also added the `-r` flag to `read`. Apart from those issues there were already a few benchmarks that were using the `-r` flag: ``` ❯ rg 'while read -r' elasticsearch/run.sh 5:cat 'queries.sql' | while read -r QUERY; do starrocks/run.sh 8:cat queries.sql | while read -r query; do doris/run.sh 8:while read -r query; do crunchy-bridge-for-analytics/run.sh 22: while read -r query; do ```
The main changes in this PR include:
\character correctly and this mistake didn't be realized by us before, sorry for that.