[Rspamd-Users] neural learning

[Ralph Meyer]

>> Hello,
>> 
>> one question. After changing some config a "systemctl reload rspamd" always
>> means,
>> loosing all learned neural data ?
> 
> Not in 2.0+


Sorry my fault.

Version: 2.1-1~xenial

After "systemctl reload rspamd" I found :

2019-10-31 12:08:48 #2877(main) rspamd_crash_sig_handler: caught fatal signal 11(Segmentation fault), pid: 2877, trace: 
2019-10-31 12:08:48 #2877(main) rspamd_print_crash: 0: 000000000048FD1F: nallocm()+0x12b3f
2019-10-31 12:08:48 #2877(main) rspamd_print_crash: 1: 0000000000477A5D: free()+0x31d
2019-10-31 12:08:48 #2877(main) rspamd_print_crash: 2: 00007F683EAB939F: corei7_hs_free_database()+0x1f
2019-10-31 12:08:48 #2877(main) rspamd_print_crash: 3: 00007F683E723639: rspamd_multipattern_destroy()+0x79
2019-10-31 12:08:48 #2877(main) rspamd_print_crash: 4: 00007F683E771A81: rspamd_url_init()+0x41
2019-10-31 12:08:48 #2877(main) rspamd_print_crash: 5: 00007F683E726592: rspamd_config_post_load()+0x322
2019-10-31 12:08:48 #2877(main) rspamd_print_crash: 6: 000000000042276D: _init()+0x10485
2019-10-31 12:08:48 #2877(main) rspamd_print_crash: 7: 00007F683D5B2F71: ev_invoke_pending()+0x71
2019-10-31 12:08:48 #2877(main) rspamd_print_crash: 8: 00007F683D5B3C4C: ev_run()+0xcbc
2019-10-31 12:08:48 #2877(main) rspamd_print_crash: 9: 0000000000422156: _init()+0xfe6e
2019-10-31 12:08:48 #2877(main) rspamd_print_crash: 10: 00007F683F2DA830: __libc_start_main()+0xf0
2019-10-31 12:08:48 #2877(main) rspamd_print_crash: 11: 00000000004142D9: _init()+0x1ff1
2019-10-31 12:08:48 #2877(main) rspamd_print_crash: 12: 0000000000000000: <unknown>
2019-10-31 12:08:48 #2888(fuzzy) <b44d45>; main; rspamd_worker_term_handler: terminating after receiving signal Terminated
2019-10-31 12:08:48 #2887(normal) <b44d45>; main; rspamd_worker_term_handler: terminating after receiving signal Terminated
2019-10-31 12:08:48 #2885(controller) <b44d45>; main; rspamd_worker_term_handler: terminating after receiving signal Terminated
2019-10-31 12:08:48 #2887(normal) rspamd_worker_on_terminate: performing finishing actions
2019-10-31 12:08:48 #2886(normal) <b44d45>; main; rspamd_worker_term_handler: terminating after receiving signal Terminated
2019-10-31 12:08:48 #2891(fuzzy) <b44d45>; main; rspamd_worker_term_handler: terminating after receiving signal Terminated
2019-10-31 12:08:48 #2890(fuzzy) <b44d45>; main; rspamd_worker_term_handler: terminating after receiving signal Terminated
2019-10-31 12:08:48 #2892(hs_helper) <b44d45>; main; rspamd_worker_term_handler: terminating after receiving signal Terminated
2019-10-31 12:08:48 #2884(rspamd_proxy) <b44d45>; main; rspamd_worker_term_handler: terminating after receiving signal Terminated
2019-10-31 12:08:48 #2889(fuzzy) <b44d45>; main; rspamd_worker_term_handler: terminating after receiving signal Terminated
2019-10-31 12:08:48 #2886(normal) rspamd_worker_on_terminate: performing finishing actions
2019-10-31 12:08:48 #2885(controller) rspamd_controller_on_terminate: closing rrd file: /var/lib/rspamd/rspamd.rrd
2019-10-31 12:08:49 #3610(main) rspamd_logger_configure_modules: enable debugging for module neural (36)
2019-10-31 12:08:49 #3610(main) <>; lua; lua_util.lua:964: enable debug for Lua module neural
2019-10-31 12:08:49 #3610(main) lua; lua_cfg_transform.lua:161: group excessqp has no symbols
2019-10-31 12:08:49 #3610(main) lua; lua_cfg_transform.lua:161: group excessb64 has no symbols
2019-10-31 12:08:49 #3610(main) <>; lua; lua_cfg_transform.lua:467: enable `options.check_all_filters` for neural network


After that "our version" and "remote version" is 0 for all configured networks.

local.d/neural.conf

servers = 127.0.0.1:6379;
enabled = true;
dbname = "2"; # Redis setup

rules {
  "NEURAL_WEEK_500" {
    train {
      max_trains = 500;
      max_usages = 50;
      max_iterations = 25;
      learning_rate = 0.01,
#      spam_score = 8;
#      ham_score = -2;
    }
    symbol_spam = "NEURAL_WEEK_SPAM";
    symbol_ham = "NEURAL_WEEK_HAM";
    ann_expire = 100d;
  }
  "NEURAL_DAYS_200" {
    train {
      max_trains = 200;
      max_usages = 10;
      max_iterations = 25;
      learning_rate = 0.01,
 #     spam_score = 8;
 #     ham_score = -2;
    }
    symbol_spam = "NEURAL_DAYS_SPAM";
    symbol_ham = "NEURAL_DAYS_HAM";
    ann_expire = 10d;
  }
  "NEURAL_HALF_DAY_50" {
    train {
      max_trains = 50;
      max_usages = 4;
      max_iterations = 25;
      learning_rate = 0.01,
 #     spam_score = 8;
 #     ham_score = -2;
    }
    symbol_spam = "NEURAL_HALF_DAY_SPAM";
    symbol_ham = "NEURAL_HALF_DAY_HAM";
    ann_expire = 3d;
  }
}

Ralph