Missing attribute 'fullpath' for evaluating boolean expression

We've been investigating what could cause the following errors for existing files when using post_sched_match = auto_update; (the default):

2022/08/05 10:31:05 [30914/3] Policy | Missing attribute 'fullpath' for evaluating boolean expression on [0x2000520e0:0x2a62:0x0]
2022/08/05 10:31:05 [30914/3] Policy | [0x2000520e0:0x2a62:0x0]: attribute is missing for checking ignore_fileclass rule
2022/08/05 10:31:05 [30914/3] checkdv | Warning: cannot determine if entry  is whitelisted: skipping it.

This has also been reported by others on the mailing list I believe.

When running the policy, we have a few occurrences per minute, so it's not insignificant.

We checked the errors with multiple FIDs, and every time, the path is OK in the DB, like for this one:

MariaDB [robinhood_fir]> SELECT this_path(parent_id,name) FROM ENTRIES LEFT JOIN NAMES ON ENTRIES.id=NAMES.id WHERE ENTRIES.id='0x2000520e0:0x2a62:0x0';
+---------------------------------------------------------------------------------------------------+
| this_path(parent_id,name)                                                                         |
+---------------------------------------------------------------------------------------------------+
| 0x200000007:0x1:0x0/users/oparedes/Orca_Calc/Geom_opt/Arun_Kummar/A2/CL20_GeomOpt.proc6.orho0.tmp |
+---------------------------------------------------------------------------------------------------+
1 row in set (0.001 sec)

However, full logs show that the path is not resolved when the policy is run (example with another FID 0x2000523b3:0x1a:0x0):

2022/08/05 09:37:08 [28176/21] ListMgr | SQL query: SELECT id FROM ENTRIES WHERE id='0x2000523b3:0x1a:0x0'
2022/08/05 09:37:08 [28176/11] ListMgr | SQL query: COMMIT
2022/08/05 09:37:08 [28176/21] checkdv | requests: OK + in flight = 14
2022/08/05 09:37:08 [28176/26] checkdv | Checking if entry  matches policy rules (mode=auto_update_attrs)
2022/08/05 09:37:08 [28176/26] checkdv | Updating info about [0x2000523b3:0x1a:0x0]
2022/08/05 09:37:08 [28176/26] checkdv | Updating POSIX info of [0x2000523b3:0x1a:0x0]
2022/08/05 09:37:08 [28176/21] ListMgr | SQL query: SELECT size,last_mod,type,checkdv_lstchk,checkdv_out,parent_id,name,path_update,this_path(parent_id,name) FROM ENTRIES LEFT JOIN NAMES ON ENTRIES.id=NAMES.id WHERE ENTRIES.id='0x2000523b3:0x1b:0x0'
2022/08/05 09:37:08 [28176/26] Policy | Missing attribute 'fullpath' for evaluating boolean expression on [0x2000523b3:0x1a:0x0]
2022/08/05 09:37:08 [28176/26] Policy | [0x2000523b3:0x1a:0x0]: attribute is missing for checking ignore_fileclass rule
2022/08/05 09:37:08 [28176/26] checkdv | Warning: cannot determine if entry  is whitelisted: skipping it.

It looks like if we set post_sched_match = force_update;, the error goes away. It's our current workaround, but I'm worried it will slow down the policy (testing now...).

I'm suspecting an issue in src/policies/policy_run.c:check_entry(), when the path is updated:

2270     /* get fullpath or name, if they are needed to apply the policy */
2271     if (need_update(check_method, updt_mask.std &
2272                         (ATTR_MASK_fullpath | ATTR_MASK_name))) {
2273         DisplayLog(LVL_FULL, tag(policy), "Updating path info of "DFID,
2274                    PFID(&p_item->entry_id));
2275         switch (path_check_update(&p_item->entry_id, stat_path, new_attr_set,
2276                                   updt_mask)) {
2277         case PCR_UPDATED:
2278             updated = true;
2279             break;
2280 
2281         case PCR_NO_CHANGE:
2282             break;
2283 
2284         case PCR_ORPHAN:
2285             /* no path to access it, handle it as if it had been moved */
2286             return AS_MOVED;
2287         }
2288     }

Any idea? :) Thx

Apologies for the delay. A policy basically looks either like this:

define_policy checkdv {
    status_manager = checker;
    scope { type == file }
    default_lru_sort_attr = none;
    # 'output' stands for previous value in DB
    # 7862400 = 90 days + 1 day grace
    default_action = cmd("/usr/sbin/rbh_checkdv /fir '{creation_time}' '{output}' 7862400 '{fid}'");
}

checkdv_rules {
    # ignore system files
    ignore_fileclass = system;

    rule default {
        condition { (checkdv.last_check == 0 or checkdv.output == "") and creation > 2d and
                    (ost_index == 0 or
                     ost_index == 1 or
                     ost_index == 2 or
                     ost_index == 3 or
                     ost_index == 4 or
                     ost_index == 5 or
                     ost_index == 6 or
                     ost_index == 7 or
                     ost_index == 8 or
                     ost_index == 9 or
                     ost_index == 10 or
                     ost_index == 11) }
    }
}

checkdv_parameters {
    db_result_size_max = 262144;
    queue_size = 65536;
    nb_threads = 8;
    reschedule_delay_ms = 0;
    recheck_ignored_entries = no;
    report_interval = 1min;
    pre_sched_match = none;
    post_sched_match = force_update;
}

...or like that:

define_policy checkdv {
    status_manager = checker;
    scope { type == file }
    default_lru_sort_attr = creation; #oldest first
    # 'output' stands for previous value in DB
    # 7862400 = 90 days + 1 day grace
    default_action = cmd("/usr/sbin/rbh_checkdv /fir '{creation_time}' '{output}' 7862400 '{fid}'");
}

checkdv_rules {
    # ignore system files
    ignore_fileclass = system;

    rule default {
        condition { checkdv.output != "" and creation > 90d and checkdv.last_check > 2d and
                    (ost_index == 0 or
                     ost_index == 1 or
                     ost_index == 2 or
                     ost_index == 3 or
                     ost_index == 4 or
                     ost_index == 5 or
                     ost_index == 6 or
                     ost_index == 7 or
                     ost_index == 8 or
                     ost_index == 9 or
                     ost_index == 10 or
                     ost_index == 11) }
    }
}

checkdv_parameters {
    db_result_size_max = 1000000000;  #no pagination? if not big enough we might never reach some entries
    queue_size = 131072;
    nb_threads = 8;
    reschedule_delay_ms = 0;
    recheck_ignored_entries = no;
    report_interval = 1min;
    pre_sched_match = none;
    post_sched_match = force_update;
}

So I believe fullpath is only used as part of this fileclass:

FileClass system {
    definition { tree == "/fir/.*" }
}

But TBH we have made quite some changes since I opened this ticket so these differ a little bit from the original policy. I will try to test a little bit more when possible. But in any case, it seems to work fine with post_sched_match = force_update;.

cea-hpc / robinhood

Missing attribute 'fullpath' for evaluating boolean expression #125