firebird-qa/tests/bugs/gh_6976_test.py

#coding:utf-8

"""
ID:          issue-6976
ISSUE:       6976
TITLE:       Lack of proper clean up after failure to initialize shared memory
DESCRIPTION:
  Test reads content of empty database (which is created auto) and makes several checks
  by writing this content to another .fdb file but WITHOUT last <N> pages.
  We have to start cut off last N pages and verisy that every time engine will instantly
  detect damage of database (i.e. not waiting for ~110 seconds as it was before fix).
  It is useless to cut off *last* 2..3 pages because engine makes reserving of space.

  Test settings for 'starting' page to be cuted off and number of pages are:
    SKIP_BACK_FROM_LAST_PAGE;
    NUM_OF_CUTED_LAST_PAGES

  Confirmed on WI-V4.0.1.2606: needed to wait for exactly 110s after each page starting from N-3.
  Checked on 3.0.8.33501, 4.0.1.2613.
FBTEST:      bugs.gh_6976

NOTES:
    [20.07.2022] pzotov

    Bug reproduced on 4.0.1.2606 but only when it is tested just after 5.0.0.591.
    In that case firebird.log will have following messages:

        HOME-AUX2	Thu Jul 21 01:25:17 2022
        I/O error during "ReadFile" operation for file "C:/TEMP/.../TMP_GH_6976.CUTED_OFF.FDB"
        Error while trying to read from file
        [localized message can be here]: "end of file encountered"


        HOME-AUX2	Thu Jul 21 01:25:17 2022 ------------------------------------ [ 1 ]
        TPC: Cannot initialize the shared memory region (header)
        I/O error during "ReadFile" operation for file "C:/TEMP/.../TMP_GH_6976.CUTED_OFF.FDB"
        Error while trying to read from file
        [localized message can be here]: "end of file encountered"


        HOME-AUX2	Thu Jul 21 01:27:07 2022 ------------------------------------ [ 2 ] diff: 110 sec.
        Operating system call WaitForSingleObject failed. Error code 0


        HOME-AUX2	Thu Jul 21 01:27:07 2022
        TPC: Cannot initialize the shared memory region (header)
        operating system directive WaitForSingleObject failed
        [localized message can be here]: "operation completed successfully"

    NB: time interval between 01:25:17 and 01:27:07 is exactly 110 seconds.

    On recent FB 4.x and 5.x median duration for delivering error to client must be about 80 ms.
    On 4.0.1.2606 this was about 640 ms (in case when it was tested without previous 5.x test).

    For concluding whether problem exists or no, MEDIAN value of time serie is used.
    Recent FB build show that time to wait exception with expected gdscode = 335544344 must be
    less than 300 ms but it depends on ServerMode (for Classic it is about 2x more than for Super).
    Median value will be compared with THRESHOLD_FOR_MAKE_CONNECT_MS variable..

    Each attempt to connect top broken DB must bring stack with TWO gdscodes:
        isc_io_error = 335544344;
        isc_io_read_err = 335544736;

    Checked on Windows: 3.0.8.33535 (SS/CS), 4.0.1.2692 (SS/CS), 5.0.0.730
"""

import os
import subprocess
import codecs
import re
import time
from datetime import datetime as dt
from datetime import timedelta
from pathlib import Path

import pytest
from firebird.qa import *


SKIP_BACK_FROM_LAST_PAGE = 15
NUM_OF_CUTED_LAST_PAGES = 20

db = db_factory()
tmp_fdb = db_factory(filename = 'tmp_gh_6976.cuted_off.fdb', async_write = True)

act_source = python_act('db')
act_broken = python_act('tmp_fdb')

#--------------------------------------------

def median(lst):
    n = len(lst)
    s = sorted(lst)
    return (sum(s[n//2-1:n//2+1])/2.0, s[n//2])[n % 2] if n else None

#--------------------------------------------

def try_cuted_off_db(act_source, act_broken, db_page_size, db_pages_cnt, cut_off_pages_cnt):

    dbsrc = act_source.db.db_path
    dbtgt = act_broken.db.db_path
    with open(dbsrc, 'rb') as f:
        fdb_binary_content = f.read( db_page_size * (db_pages_cnt - cut_off_pages_cnt) )

    with open(dbtgt, 'wb') as w:
        w.write(fdb_binary_content)

    #msg_suffix = 'on attempt to connect to DB with missed last %d pages' % cut_off_pages_cnt

    da = dt.now()
    diff_ms = -1

    # isc_io_error = 335544344;
    # isc_io_read_err = 335544736;
    expected_gds_list = (335544344, 335544736)
    try:
        #########################
        ###   c o n n e c t   ###
        #########################
        with act_broken.db.connect():
           #with act_broken.db.connect(user = 'John', password = '123'):
           # NOTE: this *can occur.
           # In that case one need to increase SKIP_BACK_FROM_LAST_PAGE setting.
           print('UNEXPECTED: can connect to DB with missed last %d pages' % cut_off_pages_cnt)

    except Exception as e:
        db = dt.now()
        diff_ms = (db-da).seconds*1000 + (db-da).microseconds//1000
        if set( expected_gds_list ).issubset( set(e.gds_codes) ):
            pass
        else:
            print('Actual list of gdscodes:')
            print(e.gds_codes)
            print('- does not contain at least one element from following list:')
            print(expected_gds_list)
    finally:
        # We have to restore VALID content of target database
        # otherwise problem on test teardown will raise:
        with open(dbsrc, 'rb') as f:
            fdb_origin_content = f.read( db_page_size * db_pages_cnt )
        with open(dbtgt, 'wb') as w:
            w.write(fdb_origin_content)


    return diff_ms

#--------------------------------------------

@pytest.mark.version('>=3.0.8')
@pytest.mark.platform('Windows')
def test_1(act_source: Action, act_broken: Action, capsys):

    THRESHOLD_FOR_MAKE_CONNECT_MS = 250 if 'classic' in act_source.vars['server-arch'].lower() else 100

    with act_source.db.connect() as con:
        db_page_size = con.info.page_size
        db_pages_all = con.info.pages_allocated
        db_pages_cnt = con.info.size_in_pages

    ##################
    ###   L O O P  ###
    ##################
    connect_establishing_ms = []
    for i in range(SKIP_BACK_FROM_LAST_PAGE, SKIP_BACK_FROM_LAST_PAGE + NUM_OF_CUTED_LAST_PAGES, 1):
        ms = try_cuted_off_db( act_source, act_broken, db_page_size, db_pages_cnt, i)
        if ms >= 0:
            connect_establishing_ms.append( ms )

    if len(connect_establishing_ms) > 0:
        median_connect_ms = median(connect_establishing_ms)
        msg_prefix = 'Median duration of receiving an error by the client: '
        if median_connect_ms <= THRESHOLD_FOR_MAKE_CONNECT_MS:
            print(msg_prefix + 'acceptable.')
        else:
            print(msg_prefix + '/* perf_issue_tag */ POOR: %s - more than threshold: %s' % ( '{:9g}'.format(median_connect_ms), '{:9g}'.format(THRESHOLD_FOR_MAKE_CONNECT_MS)  ))
            print("Check values:" )
            for p in connect_establishing_ms:
                print(p)

    else:
        print('UNEXPECTED FINAL: ENGINE COULD NOT DETECT DAMAGE OF DATABASE.')


    expected_stdout = msg_prefix + 'acceptable.'
    act_source.expected_stdout = expected_stdout
    act_source.stdout = capsys.readouterr().out
    assert act_source.clean_stdout == act_source.clean_expected_stdout