yandex-tank maximum active instances - load-testing

I have tunned ubuntu 12.04 and YAT.
When i try to load some web resource for maximun throughput my YAT can create only 32708 active instaces. Is it a maximum?
I changed value of open file descriptors and sockets for maximum but don't have expected result.

If you have not seen yet here some recommendations:
https://yandextank.readthedocs.io/en/latest/install.html
But for more guaranteed result try these commands, especially the first one (you could use more adequate number):
echo 10000000 > /proc/sys/kernel/pid_max
sysctl -w net.ipv4.ip_local_port_range="500 65535"
echo 10000000 > /proc/sys/kernel/threads-max
echo 10000000 > /proc/sys/fs/file-max
echo 10000000 > /proc/sys/vm/max_map_count
echo 20000500 > /proc/sys/fs/nr_open

Related

Vertica's vsql.exe returns errorlevel 0 when facing ERROR 3326: Execution time exceeded run time cap

I am using vsql.exe on an external Vertica database for which I don't have any administrative access. I use some views with simple SELECT+FROM+WHERE queries.
These queries 90% of the time work just fine, but some times, randomly, I get this error:
ERROR 3326:  Execution time exceeded run time cap of 00:00:45
The strange thing is that this error can happen way after those 45 seconds, even after 3 minutes. I've been told this is related to having different resource pools, but anyway I don't want to dig into that.
The problem is that when this occurs, vsql.exe returns errorlevel 0 and there is (apparently almost) no way to know this failed.
The output of the query is stored in a csv file. When it succeeds, it ends with (#### rows). But when it fails with this error, it just stops at any point of the csv, and its resulting size is around half of what's expected. This is of course not what you would expect when an error occurs, like no output or an empty one.
If there is a connection error or if the query has syntax errors, the errorlevel is not 0, so in those cases it behaves as expected.
I've tried many things, like increasing the timeout or adding -v ON_ERROR_STOP=ON to the vsql.exe parameters, but none of that helped.
I've googled a lot and found many people having this error, but the solutions are mostly related to increasing the timeouts, not related to the errorlevel returned.
Any help will be greatly appreciated.
TL;DR: how can I detect an error 3326 in a batch file like this?
#echo off
vsql.exe -h <hostname> -U <user> -w <pwd> -o output.cs -Ac "SELECT ....;"
echo %errorlevel% is always 0
if errorlevel 1 echo Error!! But this is never displayed.
Now that's really unexpected to me. I don't have Windows available just now, but trying on my Mac - at first just triggering a deliberate error:
$ vsql -h zbook -d sbx -U dbadmin -w $VSQL_PASSWORD -v ON_ERROR_STOP=ON -Ac "select * from foobarfoo"
ERROR 4566: Relation "foobarfoo" does not exist
$ echo $?
1
With ON_ERROR_STOP set to ON, this should be the behaviour everywhere.
Could you try what I did above through Windows, just with echo %ERRORLEVEL% instead of echo $?, just from the Windows command prompt and not in a batch file?
Next test: I run on resource pool general in my little test database, so I temporarily modify it to a runtime cap of 30 sec, run a silly query that will take over 30 seconds with ON_ERROR_STOP set to ON, collect the value returned by vsql and set the runtime cap of general back to NONE. I also have the %VSQL_* % env variables set so I don't have to repeat them all the time:
rem Windows way to set environment variables for vsql:
set VSQL_HOST=zbook
set VSQL_DATABASE=sbx
set VSQL_USER=dbadmin
set VSQL_PASSWORD=***masked***
Now for the test (backslashes, in Linux/MacOs escape a new line, which enables you to "word wrap" a shell command. Use the caret (^) in Windows for that):
marco ~/1/Vertica/supp $ # set a runtime cap
marco ~/1/Vertica/supp $ vsql -i -c \
"alter resource pool general runtimecap '00:00:30'"
ALTER RESOURCE POOL
Time: First fetch (0 rows): 116.326 ms. All rows formatted: 116.730 ms
marco ~/1/Vertica/supp $ vsql -v ON_ERROR_STOP=ON -iAc \
"select count(*) from one_million_rows a cross join one_million_rows b"
ERROR 3326: Execution time exceeded run time cap of 00:00:30
marco ~/1/Vertica/supp $ # test the return code
marco ~/1/Vertica/supp $ echo $?
1
marco ~/1/Vertica/supp $ # clear the runtime cap
marco ~/1/Vertica/supp $ vsql -i -c \
"alter resource pool general runtimecap NONE "
ALTER RESOURCE POOL
Time: First fetch (0 rows): 11.148 ms. All rows formatted: 11.383 ms
So it works in my case. Your line:
if errorlevel 1 echo Error!! But this is never displayed.
... never echoes anything because the previous line, with echo will return 0 to the shell, overriding the previous errorlevel.
Try it command by command on your Windows command prompt, and see what happens. Just echo %errorlevel%, without evaluating it.
And I notice that you are trying to export to CSV format. Then, try this:
Format the output unaligned (-A)
set the field separator to comma (-F ',')
remove the footer '(n rows)' (-P footer)
limit the output to 5 rows in the query for test
(I show the output before redirecting to file):
marco ~/1/Vertica/supp $ vsql -A -F ',' -P footer -c "select * from one_million_rows limit 5"
id,id_desc,dob,category,busid,revenue
0,0,1950-01-01,1,====== boss ========,0.000
1,-1,1950-01-02,2,kbv-000001kbv-000001,0.010
2,-2,1950-01-03,3,kbv-000002kbv-000002,0.020
3,-3,1950-01-04,4,kbv-000003kbv-000003,0.030
4,-4,1950-01-05,5,kbv-000004kbv-000004,0.040
Not aligning is much faster than aligning.
Then, as you spend most time in the fetching of the rows (that's because you get a timeout in the middle of an output file write process), try fetching more rows at a time than the default 1000. You will need to play with the value, depending on the network settings at your site until you get your best value:
-v ROWS_AT_A_TIME=10000
Once you're happy with the tested output, try this command (change the SELECT for your needs, of course ....):
marco ~/1/Vertica/supp $ vsql -A -F ',' -P footer \
-v ON_ERROR_STOP=ON -v ROWS_AT_A_TIME=10000 -o one_million_rows.csv \
-c "select * from one_million_rows"
marco ~/1/Vertica/supp $ wc -l one_million_rows.csv
1000001 one_million_rows.csv
The table actually contains one million rows. Note the line count in the file: 1,000,001. That's the title line included, but the footer (1000000 rows) removed.

parallel: Error: Command line too long (68914 >= 65524) at input 0

Given a file with long lines, parallel fails to pass these lines as an argument to any command:
$> cat johny_long_lines.txt | parallel echo {}
parallel: Error: Command line too long (68906 >= 65524) at input 0: 2236439425|\x308286873082856fa003020102020c221ff03...
This gets more confusing when I see that the line is 68900 characters long:
$> cat johny_long_lines.txt | head -n 1 | wc -m
68900
while the max line length allowed by parallel is way more longer that my input:
$> parallel --max-line-length-allowed
131049
Also: if you think that it's a problem of execve, this might interest you:
$> getconf ARG_MAX
2097152
Any idea what I'm doing here wrong?
UPDATE
I figured out that the problem persists for versions 20161222 and 20220522 but not for 20210822 (delivered with Ubuntu 22.04 LTS). Further inspection reveals that this line causes the problem:
# Usable len = maxlen - 3000 for wrapping, div 2 for hexing
int(($Global::minimal_command_line_length - 3000)/2);
Which I can confirm using --show-limits:
$> parallel --show-limits
[...]
Maximal size of command: 131063
Maximal usable size of command: 64031
This annoying feature does not exist in version 20210822 and I my file goes through as expected.
Can this be disabled?
I got the message
parallel --show-limits
Maximal size of command: 131049
Maximal used size of command: 83906
and had some trouble finding the source for the 83906
but then found it in
~/.parallel/tmp/sshlogin/$(hostname)/linelen
and had no clue how it was set to this small value.
But the Version of parallel was quite old:
GNU parallel 20180922

check_cpu + nsclient : set critical threshold only on 5min period

I am using centreon (nagios) to monitor the CPUs of some VMs using NSClient. In my case it makes only sense to set the critical state of the cpu probe if the average cpu load is > 95 over the 5m period. Is this achievable ?
I cannot find documentation on how to specify that in the critical param
Default command
check_cpu
Returns
CPU Load ok
'total 5m load'=0%;80;90 'total 1m load'=0%;80;90 'total 5s load'=7%;80;90
Command with specific threshold (but all time period can match)
check_cpu "critical=load > 90"
It is not exactly what I wanted to do but what I did is the following
check_nrpe -u -H XX.XXX.X.XXX -c check_cpu -a "crit=load > 95" "warn=load > 90" time=5m
Which limits the output to the 5m time period.
Note that to execute this from centreon you have to set the following variables inside the nsclient.ini file (waisted a lot of time on that one)
[/settings/NRPE/server]
allow nasty characters=true
[/settings/external scripts]
allow nasty characters=true
Check this script,
define service{
use generic-service
host_name xxx
service_description CPU Load
check_command check_nrpe!check_load
contact_groups sysadmin
}
---
command[check_load]=/usr/local/nagios/libexec/check_load -w 15,10,5 -c 30,25,20
You can try something like that
check_nrpe -u -H XX.XXX.X.XXX -c check_cpu -a "warning=time = '5m' and load > 80" "critical=time = '5m' and load > 90" show-all
You can also check the documentation for more info.

Snakemake memory limiting

In Snakemake, I have 5 rules. For each I set the memory limit by resources mem_mb option.
It looks like this:
rule assembly:
input:
file1 = os.path.join(MAIN_DIR, "1.txt"), \
file2 = os.path.join(MAIN_DIR, "2.txt"), \
file3 = os.path.join(MAIN_DIR, "3.txt")
output:
foldr = dir, \
file4 = os.path.join(dir, "A.png"), \
file5 = os.path.join(dir, "A.tsv")
resources:
mem_mb=100000
shell:
" pythonscript.py -i {input.file1} -v {input.file2} -q {input.file3} --cores 5 -o {output.foldr} "
I want to limit the memory usage of the whole Snakefile by doing something like:
snakamake --snakefile mysnakefile_snakefile --resources mem_mb=100000
So not all jobs would use 100GB each ( if I have 5 rules, meaning as 500GB memory allocation), but all of their executions will be maximum 100GB ( 5 jobs, total of 100 GB allocation?)
The command line argument sets the total limit. The Snakemake scheduler will ensure that for the set of running jobs, the sum of the mem_mb resources will not exceed the total limit.
I think this is exactly what you want, isn't it? You just need to set the per-job expected memory in the rule itself. Note that Snakemake does not measure this for you. You have to define that value yourself in the rule. E.g., if you expect your job to use 100MB memory, put mem_mb=100 into that rule.

Curl a URL list from a file and make it faster with parallel

Right now i'm using the followin code:
while read num;
do M=$(curl "myurl/$num")
echo "$M"
done < s.txt
where s.txt contains a list (1 per line) of a part of the url.
Is it correct to assume that curl is running sequentially?
Or is it running in thread/jobs/multiple conn at a time?
I've found this online:
parallel -k curl -s "http://example.com/locations/city?limit=100\&offset={}" ::: $(seq 100 100 30000) > out.txt
The problem is that my sequence is coming from a file or from a variable (one element per line) and i can't adapt it to my needs
I've not fully understood how to pass the list to parallel
Should i save all the curl commands in the list and run it with parallel -a ?
Regards,
parallel -j100 -k curl myurl/{} < s.txt
Consider spending an hour walking through man parallel_tutorial. Your command line will love you for it.

Resources