TCXB8-4164: Re-enable profile scheduler when hash-match skips dead thread - #410

source/bulkdata/reportprofiles.c:1000

• Calling enableProfile(profileName) here is unlikely to restart a dead scheduler thread: enableProfile() returns early when profile->enable is already true, and registerProfileWithScheduler() also exits when a scheduler entry already exists for the profile name. If the existing scheduler entry’s TimeoutThread died silently, this path will still leave the profile unscheduled. A dedicated scheduler restart path (or making registerProfileWithScheduler recreate the thread when the existing one is not alive) seems necessary to meet the PR description.

                if(!isProfileSchedulerRunning(profileName))
                {
                    T2Warning("Profile %s scheduler not running, re-enabling\n", profileName);
                    enableProfile(profileName);
                }

source/bulkdata/reportprofiles.c:1415

• Same as the JSON path above: enableProfile(profileName) will no-op when the profile is already enabled, and registerProfileWithScheduler() returns success when an entry already exists, so this is unlikely to restart a dead TimeoutThread. Consider adding a scheduler "restart" API or updating registerProfileWithScheduler() to recreate the thread when the existing one is not alive.

                if(!isProfileSchedulerRunning(profileName))
                {
                    T2Warning("Profile %s scheduler not running, re-enabling\n", profileName);
                    enableProfile(profileName);
                }

source/bulkdata/reportprofiles.c:1415

• Same issue as the JSON path: enableProfile(profileName) will not restart scheduling if the profile is already enabled (enableProfile() returns early when profile->enable is true; see source/bulkdata/profile.c:1039-1044). This means the intended recovery for ‘hash match but dead scheduler thread’ will not occur.

                if(!isProfileSchedulerRunning(profileName))
                {
                    T2Warning("Profile %s scheduler not running, re-enabling\n", profileName);
                    enableProfile(profileName);
                }

source/bulkdata/reportprofiles.c:1000

• The new ‘hash matches but scheduler thread is not running’ recovery path is not covered by unit tests. There is existing bulkdata unit test coverage for ReportProfiles_ProcessReportProfilesBlob / msgpack processing (source/test/bulkdata/reportprofilesTest.cpp), but no test exercises the hash-match branch with isProfileSchedulerRunning() returning false and verifies the expected recovery action is taken.

                if(!isProfileSchedulerRunning(profileName))
                {
                    T2Warning("Profile %s scheduler not running, re-enabling\n", profileName);
                    enableProfile(profileName);
                }

source/bulkdata/reportprofiles.c:1000

• Calling enableProfile(profileName) here will not restart a dead scheduler thread when the Profile is already marked enabled. enableProfile() returns early for enabled profiles (source/bulkdata/profile.c:1039-1044), so this new recovery path is effectively a no-op in the failure scenario described in the PR (thread init failure but profile stays enabled). Consider adding an explicit ‘restart scheduler’ path (e.g., force re-register with scheduler when enabled but scheduler not running) instead of calling enableProfile unconditionally.

This issue also appears in the following locations of the same file:

• line 996
• line 1411

                if(!isProfileSchedulerRunning(profileName))
                {
                    T2Warning("Profile %s scheduler not running, re-enabling\n", profileName);
                    enableProfile(profileName);
                }

source/scheduler/scheduler.c:453

• isProfileSchedulerRunning() uses pthread_kill(tProfile->tId, 0) to test liveness, but scheduler.c currently ignores pthread_create() failures when setting tProfile->tId (registerProfileWithScheduler() calls pthread_create without checking its return). In the PR’s described failure mode (thread creation/init failure), tProfile->tId may be left uninitialized/invalid, and calling pthread_kill on an invalid pthread_t is undefined on some pthread implementations. Consider recording thread-creation success (or only storing profiles in profileList after a successful pthread_create) so this liveness check is safe and reliable.

bool isProfileSchedulerRunning(const char* profileName)
{
    if(!sc_initialized || profileName == NULL || profileList == NULL)
        return false;

    if(pthread_mutex_lock(&scMutex) != 0)
        return false;

    size_t index = 0;
    for(; index < profileList->count; ++index)
    {
        SchedulerProfile *tProfile = (SchedulerProfile *)Vector_At(profileList, index);
        if(tProfile == NULL || tProfile->name == NULL)
            continue;
        if(strcmp(tProfile->name, profileName) == 0)
        {
            /* pthread_kill with signal 0 checks if the thread is still alive */
            bool running = (pthread_kill(tProfile->tId, 0) == 0);
            pthread_mutex_unlock(&scMutex);
            return running;