ObserverWeb.Apps.Aggregator (Observer Web v0.2.7)

View Source

Aggregates an application's supervision tree (as built by ObserverWeb.Apps.info/2) into per-application totals - the observer_cli App pane's view: how many processes/ports an application holds and how much memory, reductions and message-queue backlog they account for.

Counting walks the already-fetched tree, so it costs no RPCs. Summing process stats does one Rpc.pinfo/2 per process (location transparent, so remote applications work), which is why it is a separate, on-demand step - and why it's capped: trees with more processes than the cap report a partial sum, flagged as such, instead of hammering the observed node.

Summary

Types

counts()
stats()

Functions

count(tree)

Counts the tree's processes, ports and references. Pure tree walk, no RPCs.

stats(tree)

Sums memory, reductions and message-queue length over the tree's processes (one Rpc.pinfo/2 per process, capped at 2000 - see the moduledoc). Processes that died since the tree was built are skipped.

Types

counts()

@type counts() :: %{
  processes: non_neg_integer(),
  ports: non_neg_integer(),
  references: non_neg_integer()
}

stats()

@type stats() :: %{
  memory: non_neg_integer(),
  reductions: non_neg_integer(),
  message_queue_len: non_neg_integer(),
  sampled: non_neg_integer(),
  partial?: boolean()
}

Functions

count(tree)

@spec count(ObserverWeb.Apps.t() | map()) :: counts()

Counts the tree's processes, ports and references. Pure tree walk, no RPCs.

stats(tree)

@spec stats(ObserverWeb.Apps.t() | map()) :: stats()

Sums memory, reductions and message-queue length over the tree's processes (one Rpc.pinfo/2 per process, capped at 2000 - see the moduledoc). Processes that died since the tree was built are skipped.