ClickHouse/tests/queries/0_stateless/01658_read_file_to_stringcolumn.sh
Ivan 57f61c954c
Another attempt to enable pytest (#22664)
* Fix some tests

* Fix tests
2021-04-14 19:35:17 +03:00

90 lines
4.0 KiB
Bash
Executable File

#!/usr/bin/env bash
set -eu
CURDIR=$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)
# shellcheck source=../shell_config.sh
. "$CURDIR"/../shell_config.sh
# Data preparation.
# Now we can get the user_files_path by use the table file function for trick. also we can get it by query as:
# "insert into function file('exist.txt', 'CSV', 'val1 char') values ('aaaa'); select _path from file('exist.txt', 'CSV', 'val1 char')"
user_files_path=$(clickhouse-client --query "select _path,_file from file('nonexist.txt', 'CSV', 'val1 char')" 2>&1 | grep Exception | awk '{gsub("/nonexist.txt","",$9); print $9}')
mkdir -p ${user_files_path}/
echo -n aaaaaaaaa > ${user_files_path}/a.txt
echo -n bbbbbbbbb > ${user_files_path}/b.txt
echo -n ccccccccc > ${user_files_path}/c.txt
echo -n ccccccccc > /tmp/c.txt
mkdir -p ${user_files_path}/dir
### 1st TEST in CLIENT mode.
${CLICKHOUSE_CLIENT} --query "drop table if exists data;"
${CLICKHOUSE_CLIENT} --query "create table data (A String, B String) engine=MergeTree() order by A;"
# Valid cases:
${CLICKHOUSE_CLIENT} --query "select file('${user_files_path}/a.txt'), file('${user_files_path}/b.txt');";echo ":"$?
${CLICKHOUSE_CLIENT} --query "insert into data select file('${user_files_path}/a.txt'), file('${user_files_path}/b.txt');";echo ":"$?
${CLICKHOUSE_CLIENT} --query "insert into data select file('${user_files_path}/a.txt'), file('${user_files_path}/b.txt');";echo ":"$?
${CLICKHOUSE_CLIENT} --query "select file('${user_files_path}/c.txt'), * from data";echo ":"$?
${CLICKHOUSE_CLIENT} --multiquery --query "
create table filenames(name String) engine=MergeTree() order by tuple();
insert into filenames values ('a.txt'), ('b.txt'), ('c.txt');
select file(name) from filenames format TSV;
drop table if exists filenames;
"
# Invalid cases: (Here using sub-shell to catch exception avoiding the test quit)
# Test non-exists file
echo "clickhouse-client --query "'"select file('"'nonexist.txt'), file('${user_files_path}/b.txt')"'";echo :$?' | bash 2>/dev/null
# Test isDir
echo "clickhouse-client --query "'"select file('"'${user_files_path}/dir'), file('${user_files_path}/b.txt')"'";echo :$?' | bash 2>/dev/null
# Test path out of the user_files directory. It's not allowed in client mode
echo "clickhouse-client --query "'"select file('"'/tmp/c.txt'), file('${user_files_path}/b.txt')"'";echo :$?' | bash 2>/dev/null
# Test relative path consists of ".." whose absolute path is out of the user_files directory.
echo "clickhouse-client --query "'"select file('"'${user_files_path}/../../../../tmp/c.txt'), file('b.txt')"'";echo :$?' | bash 2>/dev/null
echo "clickhouse-client --query "'"select file('"'../../../../a.txt'), file('${user_files_path}/b.txt')"'";echo :$?' | bash 2>/dev/null
### 2nd TEST in LOCAL mode.
echo -n aaaaaaaaa > a.txt
echo -n bbbbbbbbb > b.txt
echo -n ccccccccc > c.txt
mkdir -p dir
#Test for large files, with length : 699415
c_count=$(wc -c ${CURDIR}/01518_nullable_aggregate_states2.reference | awk '{print $1}')
echo $c_count
# Valid cases:
# The default dir is the CWD path in LOCAL mode
${CLICKHOUSE_LOCAL} --query "
drop table if exists data;
create table data (A String, B String) engine=MergeTree() order by A;
select file('a.txt'), file('b.txt');
insert into data select file('a.txt'), file('b.txt');
insert into data select file('a.txt'), file('b.txt');
select file('c.txt'), * from data;
select file('/tmp/c.txt'), * from data;
select $c_count, $c_count -length(file('${CURDIR}/01518_nullable_aggregate_states2.reference'))
"
echo ":"$?
# Invalid cases: (Here using sub-shell to catch exception avoiding the test quit)
# Test non-exists file
echo "clickhouse-local --query "'"select file('"'nonexist.txt'), file('b.txt')"'";echo :$?' | bash 2>/dev/null
# Test isDir
echo "clickhouse-local --query "'"select file('"'dir'), file('b.txt')"'";echo :$?' | bash 2>/dev/null
# Restore
rm -rf a.txt b.txt c.txt dir
rm -rf ${user_files_path}/a.txt
rm -rf ${user_files_path}/b.txt
rm -rf ${user_files_path}/c.txt
rm -rf /tmp/c.txt
rm -rf ${user_files_path}/dir