bernisys
commented
3 years ago We still get graphs containing duplicated items, so i did some further investigation ...
As an example, this is how the graphs look like:
You see the legend-items all doubled, also the graphs are plotted twice. so I checked the DB once more:
MariaDB [cacti]> select local_graph_id, count(1) from graph_templates_item where graph_template_id=38 group by local_graph_id limit 10;
+----------------+----------+
| local_graph_id | count(1) |
+----------------+----------+
| 0 | 25 | <- that's the template declaration - with local_graph_id == 0
| 230 | 25 | <-- these are all fine, 25 items in the graph, like in the template
| 231 | 25 |
| 232 | 25 |
| 233 | 25 |
| 234 | 25 |
| 235 | 25 |
| 236 | 25 |
| 237 | 25 |
| 238 | 25 |
+----------------+----------+Then i searched for devices with different graph item count and found this:
MariaDB [cacti]> select local_graph_id, items from (select local_graph_id, count(1) as items from graph_templates_item where graph_template_id=38 group by local_graph_id) as result where items != 25 limit 10;
+----------------+-------+
| local_graph_id | items |
+----------------+-------+
| 65757 | 22 | <-- too small item count
| 184054 | 3 | <-- way too small
| 200848 | 14 | <-- *sigh*
| 448170 | 13 | <-- what happens here?!
| 547412 | 50 | <-- aaand duplicates
| 547420 | 50 |
| 547422 | 50 |
| 547430 | 50 |
| 547432 | 50 |
| 547441 | 50 |
+----------------+-------+65757 is a graph my colleague remembers, he did a device-cleanup recently (i.e. removing old graphs and re-discover the device)
547412 and following are the graphs from devices he re-discovered recently during the cleanup
All the other like 250k traffic graphs are fully functional and look okay:
MariaDB [cacti]> select items, count(1) from (select local_graph_id, count(1) as items from graph_templates_item where graph_template_id=38 group by local_graph_id) as result group by items;
+-------+----------+
| items | count(1) |
+-------+----------+
| 3 | 1 |
| 13 | 1 |
| 14 | 1 |
| 22 | 2 |
| 25 | 245375 | <- major amount of graphs is okay
| 48 | 2 |
| 49 | 1 |
| 50 | 4424 |
+-------+----------+Checking the graphs-table:
MariaDB [cacti]> select id, local_graph_template_graph_id as LGTGI, local_graph_id, graph_template_id as GTI, title_cache from graph_templates_graph where graph_template_id = 38 and local_graph_id in (65757, 184054, 200848, 448170, 547412);
+--------+-------+----------------+-----+------------------------------------------------------------------------------+
| id | LGTGI | local_graph_id | GTI | title_cache |
+--------+-------+----------------+-----+------------------------------------------------------------------------------+
| 65895 | 186 | 65757 | 38 | xxxxxxxxxxx10-14 - Interface - Ethernet100/1/7 Traffic (xxxxxxxx NIC1_eth1) |
| 184189 | 186 | 184054 | 38 | xxxxxxxxxxxx-7 - Interface - Ethernet115/1/10 Traffic (xxxxxxxx A_vmnic7) |
| 200983 | 186 | 200848 | 38 | xxxxxxxxxxxx1-o7 - Interface - Ethernet111/1/3 Traffic (xxxxxxxx NIC2_prod) |
| 448349 | 186 | 448170 | 38 | xxxxxxtest5ksb - Interface - Ethernet109/1/14 Traffic (xxxxxxxx (B&R) |
| 547612 | 186 | 547412 | 38 | xxxxxxxxxxxx10-13 - Interface - Ethernet104/1/3 Traffic (xxxxxxxx NIC1_eth1) |
+--------+-------+----------------+-----+------------------------------------------------------------------------------+
5 rows in set (0.00 sec)So the graphs look valid, but nothing is displayed when i point my browser to the IDs with less than 25 items ... and the 547412 one contains all duplicates - like in the image on top of this post.
Any ideas?
gj00354347
netniV
TheWitness
Describe the bug
As i understand it, a template declaration inside the DB is identified by the hashes being set and local_graph_template_item_id=0 and local_graph_id=0. While a real graph contains no hash and has local_graph_template_item_id pointed to the row that contains the template declaration from above and local_graph_id pointing to the graph declaration, as well a task_item_id for the elements of a graph, which are connected to a queried item.
In our DB i have found some strange duplicate entries, which relate to different local IDs than the original template. Here's an example .. watch how for example in row id=10488739 there is a duplicate of the previous row, but the local_graph_template_item_id LGTII points suddenly to a different "master ID". Instead of the correct entry 477305 it uses now 10488701, which belonged to a graph but is not even there any more, which can in turn probably cause trouble when updating a graph template and the changes are not propagated properly into all graphs due to this shift in relation.
And there are many more of those duped rows with TII=0 for the same template, with always the same pattern. I still have to check our other templates, there were a few more which showed this pattern.
And there was another problem with an apparently deleted graph of the same template, which made the template itself contain all duplicate items, because the above mentioned IDs were set to zero, hence cacti identified those entries as part of a template. I fixed that already simply by removing those problematic entries.
Fixing should be halfway doable by scanning the DB for entries with identical sequence numbers for the same local_graph_id and deleting the duplicates where the task ID is zero ... i hope at least that there is no other relation which needs to be updated in other tables. If you can you help me out on that, then i can try to create a script which scans for such entries and provide it on my github repo, so that others can use it to check their DB and correct the entries if necessary.
What i have also seen is that the entry with id=1 in the table graph_templates_graph was overwritten with some other graph template's data, this looks like a related problem - not sure if that was fixed in the meantime, you have more insight than me probably. Would be nice if someone could enlighten me here as well.
I feel that this could this be an aftermath related to the old template edit bug that was fixed in 1.2.17 - #4237
Thanks!