Changes between Initial Version and Version 2 of Ticket #1567


Ignore:
Timestamp:
Jul 10, 2018, 9:56:48 AM (6 years ago)
Author:
Dimitar Misev
Comment:

Legend:

Unmodified
Added
Removed
Modified

  • Ticket #1567

    • Property Status new → assigned
    • Property Reporter changed from bbell to Dimitar Misev
    • Property Cc Peter Baumann added
    • Property Milestone 10.0 → 9.7
    • Property Owner set to dkamov

  • Ticket #1567 – Description

    initial v2  
    1 In order to control memory leaks in Rasdaman, one should additionally pass system test queries in "test_select" with:
     1Add a `systemtest/testcases_mandatory/test_memory_leaks/test.sh` script that will run same as `test_select`, but use the following instead of rasql:
    22
    3 valgrind --leak-check=full directql 'query' -d DB_DIRECTORY
     3{{{
     4valgrind --tool=memcheck --leak-check=full --track-origins=yes directql ..
     5}}}
    46
    5 Ideally, one would parse the output of valgrind and use that data to determine whether or not a leak is excessive.
     7In the `output/` directory the output of valgrind should be saved (not the query result like in test_select). The goal of the test is to look in this output from valgrind and determine if there is ''excessive'' memory leak. The `oracle/` directory could contain the amount of maximum acceptable leaked bytes for each query.
    68
    7 An initial threshold for when a leak is "too big" could be around 50kB; however, the size of the leak might be a linear function of the size of the stored data, so perhaps a few conditionals to determine pass/failure of queries would be of use here.
     9Afterwards (in separate tickets) we can work to fix any memory leaks, and reduce the acceptable amounts in the oracle directory.