Asama

From symptom to cause, across every layer of infrastructure

Asama traces a causal chain through hosts, VMs, storage and network, and backs each link with the state it observed.

Read-only by default · No training on your data · SOC 2 Type II

How an investigation runs

Asama keeps the state of your entire infra stack as it changes, so a past window can still be queried. For every symptom it builds a list of hypotheses, tests each one against the knowledge graph and the timeline, and traces the causal chain. Hypotheses the evidence rules out across the topology are dropped.

Most incidents are hard to debug. Transient issues clear on their own, and by the time anyone logs in the evidence is gone. Cross-layer issues surface in one layer while the cause sits in another, so the obvious suspect is rarely the real one. And finding the root cause often takes people from several teams, each looking at their own tools, piecing the story together by hand.

Every symptom starts an investigation

A baseline drifts, or someone on your team asks. Either way, the symptom names the layer it surfaced in. Asama helps you find the exact cause.

  • Starts from a baseline drift or a team request
  • Symptom, layer and time window captured up front
  • No ticket triage before the investigation begins

Incoming symptom

Today · 13:44

Reports on reporting-vm are taking minutes instead of seconds.

Raised by the finance analytics team · #infra-help

Surfaced on
reporting-vm on lab-sf-30
Layer
Virtualization
Observed
Guest disk latency p99 38 ms, baseline 4 ms
Window
From 13:38, intermittent

Hypotheses are generated

Asama reads the dependency graph and the change history to produce every path that could have produced this symptom.

  • Every dependency path from the symptom traced
  • Recent changes on each path pulled in
  • Hypotheses across virtualization, kernel, network and hardware

Dependency path

reporting-vm lab-sf-30 NFS lab-sf-31 md0 RAID1

Hypotheses

6 generated · 0 tested
  • vCPU contention on lab-sf-30

    Virtualization
  • Kernel update on lab-sf-30, 3 days ago

    Kernel
  • Packet loss on the NFS path

    Network
  • Storage degradation under /dev/md0

    Hardware

+ 2 more

14 graph edges traversed · 3 changes in the last 7 days

Evidence rules paths out

Each hypothesis is tested against the timeline and the stored state. What the evidence contradicts is discarded, and the discard is shown.

  • Each hypothesis tested against the timeline and stored state
  • Every discard shown with the evidence behind it
  • Only what the evidence supports remains

Ruled out by evidence

5 of 6 discarded
  • vCPU contention on lab-sf-30 Virtualization Ruled out

    Steal time stayed at 0.4%; host CPU peaked at 31%.

  • Packet loss on the NFS path Network Ruled out

    0 drops; RTT to lab-sf-31 steady at 0.21 ms.

  • + 3 more ruled out

Consistent with the evidence

Storage degradation under /dev/md0 on lab-sf-31 Hardware Remains

Degraded, with discards from 13:36, 2 min before latency rose.

One cause, with its trail

What is left is returned with the evidence behind it: array state, active processes, load, so the reasoning can be checked rather than trusted.

  • Root cause returned with its causal chain
  • Evidence linked: array state, processes and load
  • Reasoning you can check, not just trust
reporting-vm seems slow. Why?

Cause

Degraded RAID1 /dev/md0 on lab-sf-31, under an active discard job.

Degraded md0 + discard job NFS contention reporting-vm slow

Evidence

  • /dev/md0 in state [4/3] [UUU_]

    mdstat
  • fstrim running against /dev/md0

    Processes
  • Load on lab-sf-30 peaked, now falling

    Load

Shipped recently

Know what changed, and when

Inventory → Changelog shows every hardware change across your fleet, from the last 24 hours up to 15 days.

Data you can trust

Every investigation shows how fresh its data is, and accounts for data that’s out of range.

Anomaly trends

Investigations weigh each anomaly’s trend: Worsening, Recovering, Stable or Recurring.

FAQ

See Asama on your own fleet

Bring one recurring problem your current tooling hasn’t solved. We’ll show you how Asama detects it, finds the cause, and plans the fix.