Uploaded image for project: 'Network Observability'
  1. Network Observability
  2. NETOBSERV-1676

Sampling > 1 causes disproportionate drops

    • False
    • None
    • False
    • Hide
      Known issue (1.6):
      When the Agent PacketDrop feature is enabled, and sampling is configured greater than 1, reported dropped bytes and packet ignore the sampling configuration. While this is done on purpose to not miss any drop, a side effect is that the reported proportion of drops versus non-drops becomes biased. At very high sampling rate, such as 1:1000, it's likely that almost all the traffic appears to be dropped, when looked from the console plugin. But this is only a bias.
      Fix (1.6.1):
      When the eBPF agent `PacketDrop` feature is enabled, and sampling is configured to a value greater than `1`, reported dropped bytes and dropped packets was ignoring the sampling configuration. While this was done on purpose to not miss any drops, a side effect was that the reported proportion of drops versus non-drops became biased. For example, at a very high sampling rate, such as `1:1000`, it was likely that almost all the traffic appears to be dropped when observed from the console plugin. The new behaviour is to honour the sampling configuration with dropped bytes and packets.
      Show
      Known issue (1.6): When the Agent PacketDrop feature is enabled, and sampling is configured greater than 1, reported dropped bytes and packet ignore the sampling configuration. While this is done on purpose to not miss any drop, a side effect is that the reported proportion of drops versus non-drops becomes biased. At very high sampling rate, such as 1:1000, it's likely that almost all the traffic appears to be dropped, when looked from the console plugin. But this is only a bias. Fix (1.6.1): When the eBPF agent `PacketDrop` feature is enabled, and sampling is configured to a value greater than `1`, reported dropped bytes and dropped packets was ignoring the sampling configuration. While this was done on purpose to not miss any drops, a side effect was that the reported proportion of drops versus non-drops became biased. For example, at a very high sampling rate, such as `1:1000`, it was likely that almost all the traffic appears to be dropped when observed from the console plugin. The new behaviour is to honour the sampling configuration with dropped bytes and packets.
    • Known Issue
    • NetObserv - Sprint 254

      Description of problem:

      In the agent, drops are ignoring the sampling setting, meaning that all drops stats are reported in flows. When sampling is configured (above 1), this results in disproportionate measured dropped traffic compared to non-dropped. With high sampling (e.g. 1000) it's even going to completely hide the passing traffic in the Traffic Flows view (see screenshot attached).
      
      Another issue related to drops not being sampled is that it makes sampling configuration irrelevant at some point for reducing resource footprint, when the bulk of the collected flows are drops.

      Steps to Reproduce:

      1. Configure FlowCollector with Sampling=1000 & enable drops feature
      2. Check Traffic Flows in the console plugin
      3.
      

      Actual results:

      Only drops, even when clicking on the option to show all

      Expected results:

      Have a way to see traffic, dropped/non-dropped, in the correct proportions

              mmahmoud@redhat.com Mohamed Mahmoud
              jtakvori Joel Takvorian
              Votes:
              0 Vote for this issue
              Watchers:
              4 Start watching this issue

                Created:
                Updated:
                Resolved: