BUG in RRD Graphs for Queues



  • Currently using 1.2.3RC2 Jul30 snapshot; pfsense theme (the one with the menu at the left hand side)

    I've changed the colour palette used in the PHP script and it seems that the RRD graph reflecting very odd data patterns on the Queues page.
    In this instance, it seems to reflect an unusual amount of qlanacks and almost none on qgamesdown.
    Based on the Status: Queues page, I know for a fact that the amount of lan acks was hovering at way less than 2Kb/s and Games Down hovered around 60Kb/s to 300Kb/s constantly in the few hours before this snapshot was taken.

    If you'd take a look at the snapshot, you'd realise that the RRD graph doesn't seem to reflect the Games Down queue and yet seems to be reflecting an incredible amount of Lan Acks.



  • I really have no way to explain what happened here. Changing the colors changes nothing.

    I can only imagine that some sort of traffic hit the wrong queue.



  • @databeestje:

    I really have no way to explain what happened here. Changing the colors changes nothing.

    I can only imagine that some sort of traffic hit the wrong queue.

    Oddly, the Status: Queues page reflects the data correctly.  If it's anything to go by, I don't see a lan-queues.rrd in the /var/db/rrd
    I did not touch any of the files in this directory at all though.  Just came across the missing RRD file while trying to hunt down the root of the problem.

    This is a list of the files in the rrd directory.

    ls -l /var/db/rrd

    total 2630
    -rw-r–r--  1 root  wheel  130760 Aug 18 18:22 ipsec-packets.rrd
    -rw-r--r--  1 root  wheel  130760 Aug 18 18:22 ipsec-traffic.rrd
    -rw-r--r--  1 root  wheel  130760 Aug 18 18:22 lan-packets.rrd
    -rw-r--r--  1 root  wheel  130760 Aug 18 18:22 lan-traffic.rrd
    -rw-r--r--  1 root  wheel  487424 Aug 18 18:23 system-processor.rrd
    -rw-r--r--  1 root  wheel  487424 Aug 18 18:22 system-states.rrd
    -rw-r--r--  1 root  wheel    3475 Aug 17 23:58 updaterrd.sh
    -rw-r--r--  1 root  wheel  130760 Aug 18 18:22 wan-packets.rrd
    -rw-r--r--  1 root  wheel   65672 Aug 18 18:22 wan-quality.rrd
    -rw-r--r--  1 root  wheel  391192 Aug 18 18:22 wan-queues.rrd
    -rw-r--r--  1 root  wheel  391192 Aug 18 18:22 wan-queuesdrop.rrd
    -rw-r--r--  1 root  wheel  130760 Aug 18 18:22 wan-traffic.rrd



  • It's not missing, the shaper wizard in 1.2 only works with 2 interfaces, which is basically one set of up and down queues. So making a lan queues would be the values of the wan queues reversed.

    That is why we only have a wan-queues.rrd.



  • @databeestje:

    It's not missing, the shaper wizard in 1.2 only works with 2 interfaces, which is basically one set of up and down queues. So making a lan queues would be the values of the wan queues reversed.

    That is why we only have a wan-queues.rrd.

    That would explain the "lack" of the file.  I was kind of expecting it since the other RRD graphs came in pairs.   :P

    The situation doesn't seem to resolve itself though.  Despite trying to restart the RRD backend a few times.  Now the qGamesDown and qlanacks both shadow qOthersDownH in the RRD graph.
    It (RRD graphs) is not very important to operations but I do use it for analysis and tweaking from time to time.  Now I just get laughs from observing that I have 80% more bandwidth than I should be having while looking at the RRD graphs.   :D



  • The queues graph has always been a case of hit and miss and is very problematic to get right.

    The shaper is complicated enough that it is often done wrong, resulting in rather unpredictable graphs.

    I really can't help you with this.



  • @databeestje:

    The queues graph has always been a case of hit and miss and is very problematic to get right.

    The shaper is complicated enough that it is often done wrong, resulting in rather unpredictable graphs.

    I really can't help you with this.

    It's alright.  Appreciate you guys looking into it.

    Anyway, I just figured out what the problem is.

    Instead of directly graphing qlanacks as it's absolute figure, the script/ code is actually graphing qlanacks as:  qOthersDownH - qlanacks.

    I just removed the qOthersDownH queue in the traffic shaper and now it is graphing qlanacks as:  qGamesDown - qlanacks.

    Evidently, it's just taking the last downstream queue in the line and subtracting the actual qlanacks from it for some unknown reason.

    The qLanDef don't seem to show up on the RRD graphs either but I don't quite require that on the graph.



  • If you try to graph it with the inverse style, do that values look better then?



  • @databeestje:

    If you try to graph it with the inverse style, do that values look better then?

    Nope.  If I graph it as inverse style, it takes the qGamesDown - qlanacks and charts that as qLanDef.  Seems like an odd issue to me.
    I've been able to actually replicate this at home in both 1.2.2 and 1.2.3RC2 (Jul 16 snapshot).
    Obviously, it's not related to the change of colours (as far as I can decipher from the PHP codes; not a lot but enough to figure that it only presents a colour chart to the RRDtool not unlike style classes in a CSS).

    This is what turns up in Inverse Style:



  • Just took another look at the graphs again.  I'm absolutely certain that the graph is not even charting any of the LAN queues properly.
    The alleged qGamesDown queue in the graph is simply a graph of all downstream traffic.


Log in to reply