aprs.me/lib/aprsme/telemetry/database_metrics.ex
Graham McIntire 24796f98d4
Optimize prod: scale deployment, drop expensive metric queries, fix slow callsign and nearby-stations queries
- k8s: 2 replicas, POOL_SIZE 25, cpu 2000m, mem 1Gi (was 1 replica, 5, 500m, 512Mi)
- telemetry poller 10s -> 60s; drop pg_database_size, pg_stat_statements,
  pg_table_size and pg_indexes_size (these were ~7100s of cumulative DB time
  on their own)
- get_latest_packet_for_callsign: bound by :packet_retention_days so partition
  pruning kicks in
- get_nearby_stations_knn: ST_DWithin spatial pre-filter (default 500km) so
  the GiST geography index cuts candidates before DISTINCT ON sort
  (EXPLAIN: 5793ms -> 400ms)
- New migration: partial functional indexes on upper(object_name) and
  upper(item_name), built per-partition CONCURRENTLY then attached to parent.
  Lets get_latest_packet_for_callsign use BitmapOr across all 3 upper()
  indexes instead of falling back to a scan
2026-05-12 17:48:58 -05:00

205 lines
5.7 KiB
Elixir

defmodule Aprsme.Telemetry.DatabaseMetrics do
@moduledoc """
Collects database metrics from PostgreSQL and PgBouncer
"""
require Logger
def collect_db_pool_metrics do
pool_size = Application.get_env(:aprsme, Aprsme.Repo)[:pool_size] || 10
case Process.whereis(Aprsme.Repo) do
nil -> report_pool_metrics_unavailable(pool_size)
repo_pid when is_pid(repo_pid) -> collect_from_repo(repo_pid, pool_size)
end
rescue
e ->
Logger.debug("Error collecting pool metrics: #{inspect(e)}")
report_pool_metrics_error()
end
defp collect_from_repo(repo_pid, pool_size) do
children = Supervisor.which_children(repo_pid)
pool_info = find_pool_child(children)
collect_from_pool(pool_info, pool_size)
end
defp find_pool_child(children) do
Enum.find(children, fn
{DBConnection.ConnectionPool, _, _, _} -> true
{DBConnection.Ownership, _, _, _} -> true
_ -> false
end)
end
defp collect_from_pool({_, pool_pid, _, _}, pool_size) when is_pid(pool_pid) do
# Try to get pool telemetry
case try_get_pool_telemetry(pool_size) do
{:ok, _metrics} -> report_pool_metrics_estimated(pool_size)
_ -> report_pool_metrics_idle(pool_size)
end
end
defp collect_from_pool(_, pool_size), do: report_pool_metrics_idle(pool_size)
defp try_get_pool_telemetry(pool_size) do
{:ok, %{pool_size: pool_size, idle_time: 0, queue_time: 0}}
catch
_, _ -> {:error, :not_available}
end
defp report_pool_metrics_unavailable(pool_size) do
emit_pool_metrics(%{
size: pool_size,
idle: 0,
busy: 0,
available: 0,
queue_length: 0,
total: 0
})
end
defp report_pool_metrics_idle(pool_size) do
emit_pool_metrics(%{
size: pool_size,
idle: pool_size,
busy: 0,
available: pool_size,
queue_length: 0,
total: pool_size
})
end
defp report_pool_metrics_estimated(pool_size) do
# Cannot reliably introspect DBConnection pool state, so report
# pool_size as total without guessing busy/idle breakdown.
report_pool_metrics_idle(pool_size)
end
defp report_pool_metrics_error do
emit_pool_metrics(%{
size: 0,
idle: 0,
busy: 0,
available: 0,
queue_length: 0,
total: 0
})
end
defp emit_pool_metrics(metrics) do
:telemetry.execute([:aprsme, :repo, :pool], metrics, %{})
end
def collect_postgres_metrics do
do_collect_postgres_metrics(Application.get_env(:aprsme, :env))
end
# Skip database metrics collection in the test environment — the extra
# pg_stat_* queries add test-suite noise and sometimes fail in CI sandboxes.
defp do_collect_postgres_metrics(:test), do: :ok
defp do_collect_postgres_metrics(_env) do
case Process.whereis(Aprsme.Repo) do
# Repo not started yet, skip metrics collection silently.
nil -> :ok
_pid -> collect_database_metrics()
end
end
defp collect_database_metrics do
collect_connection_stats()
collect_packets_table_stats()
collect_replication_lag()
rescue
e ->
Logger.debug("Error collecting PostgreSQL metrics: #{inspect(e)}")
end
defp collect_connection_stats do
case Aprsme.Repo.query("""
SELECT
count(*) as total,
count(*) FILTER (WHERE state = 'active') as active,
count(*) FILTER (WHERE state = 'idle') as idle,
count(*) FILTER (WHERE state = 'idle in transaction') as idle_in_transaction,
count(*) FILTER (WHERE wait_event_type IS NOT NULL) as waiting
FROM pg_stat_activity
WHERE datname = current_database()
""") do
{:ok, %{rows: [[total, active, idle, idle_in_tx, waiting]]}} ->
:telemetry.execute(
[:aprsme, :postgres, :connections],
%{
total: to_number(total),
active: to_number(active),
idle: to_number(idle),
idle_in_transaction: to_number(idle_in_tx),
waiting: to_number(waiting)
},
%{}
)
_ ->
:ok
end
end
defp collect_packets_table_stats do
case Aprsme.Repo.query("""
SELECT
n_live_tup as live_tuples,
n_dead_tup as dead_tuples,
n_tup_ins as inserts,
n_tup_upd as updates,
n_tup_del as deletes
FROM pg_stat_user_tables
WHERE relname = 'packets'
""") do
{:ok, %{rows: [[live, dead, ins, upd, del]]}} ->
emit_packets_table_telemetry(live, dead, ins, upd, del)
_ ->
:ok
end
end
defp collect_replication_lag do
case Aprsme.Repo.query("""
SELECT
extract(epoch from (now() - pg_last_xact_replay_timestamp()))::int as lag_seconds
WHERE pg_is_in_recovery()
""") do
{:ok, %{rows: [[lag]]}} when not is_nil(lag) ->
:telemetry.execute([:aprsme, :postgres, :replication], %{lag_seconds: lag}, %{})
_ ->
:ok
end
end
def collect_pgbouncer_metrics do
# PgBouncer metrics would require a separate connection to PgBouncer's admin interface
# For now, we'll skip these as they require additional setup
:ok
end
defp emit_packets_table_telemetry(live, dead, ins, upd, del) do
:telemetry.execute(
[:aprsme, :postgres, :packets_table],
%{
live_tuples: to_number(live),
dead_tuples: to_number(dead),
total_inserts: to_number(ins),
total_updates: to_number(upd),
total_deletes: to_number(del)
},
%{}
)
end
# Coerce Decimal/nil values to native Erlang numbers for :telemetry.execute
defp to_number(nil), do: 0
defp to_number(%Decimal{} = d), do: Decimal.to_float(d)
defp to_number(n) when is_number(n), do: n
defp to_number(_), do: 0
end