All 4 webhook controllers now verify the signature, insert an Oban job,
and return immediately. Previously, Stripe and PagerDuty did DB queries +
writes inline; Gaiia ran upserts; Agent Release broadcast to all connected
WebSockets — all in the request path.
New workers:
- StripeWebhookWorker — delegates to WebhookProcessor.process/1
- GaiiaWebhookWorker — delegates to Webhooks.process_event/3
- PagerdutyWebhookWorker — resolve/acknowledge alerts by dedup key
- AgentReleaseWebhookWorker — broadcast mass update to agents
Also fix: log error when resolve_alert fails in agent_channel (was silently
discarded, causing "Resolving stuck device_down alert" to repeat in logs).
- ToweropsWeb.AgentChannel now runs sensor readings through
Towerops.Snmp.SensorScale.normalize/2 in both resolve_sensor_value/2
and the GraphQL publisher. Without this, MikroTik mtxrHl* deci-degree
readings (e.g. Culleoka reporting 294 -> 29.4°C) were stored and
alerted on at 294°C. The DevicePollerWorker path already did this;
only the agent path was missing it.
- Demote "Received SNMP result from agent" and "Processing polling
result for X" from info to debug — they fire every poll cycle on
every device and were dominating prod logs.
Performance:
- schedule_live: preload page once instead of get_schedule!/1 per row
- alert_live + alerts: DB-side status filter; Repo.aggregate counts replace length/Enum.count over 500-row fetches on every event
- dashboard_live: drop duplicate get_device_status_counts; cap active alerts to 20 + use count_active_alerts/1
- maintenance.active_windows_for_device: 3 round-trips collapsed into one query (per-site-id branch keeps Postgres parameter types unambiguous)
- sites.build_site_tree: O(N^2) -> O(N) via group_by(parent_site_id)
- accounts.sole_owner_organizations: single group_by + having instead of per-org Repo.aggregate loop
- agents + agent_live (org + admin): count_assigned_devices_batch/1 + count_agent_polling_targets/1 (no preloads)
- alert_digest_worker: list_alerts_by_ids/1 batches digest fetch
- gaiia: distinct: true at DB; limit 50 on bidirectional ilike
Indexes:
- maintenance_windows(organization_id, starts_at, ends_at) WHERE suppress_alerts = true
- alerts(check_id) WHERE resolved_at IS NULL
Antipatterns:
- agents.delete_agent_token: PubSub.broadcast moved outside Repo.transaction so a rollback no longer leaves subscribers acting on a non-existent deletion
- integrations_controller.to_atom_keys: replaced String.to_existing_atom on user-controlled JSON keys with explicit allowlist
- 9 Task.start callsites converted to Task.Supervisor.start_child(Towerops.TaskSupervisor, ...) so background DB writes survive shutdown (test-mode discovery shims left as-is for sandbox semantics)
Changes to eliminate @dialyzer suppressions by fixing underlying causes:
NIF stubs (towerops_native.ex, mib_translator.ex):
- Change stubs to :erlang.nif_error(:nif_not_loaded) (no_return type).
Real NIF replaces stubs at load time; calls to unloaded stubs now fail
loudly instead of returning fake data. Lets dialyzer trust @spec.
- Remove @dialyzer :nowarn_function on three NIFs and on translate/1.
Discovery sync_* functions (snmp/discovery.ex, channels/agent_channel.ex):
- agent_channel passes %{device_id: _, interfaces: _} and %{id: _} maps
into Discovery.sync_ip_addresses/sync_processors/sync_storage, which
@spec'd only %Device{}. Add narrow map-type unions (ip_sync_device,
snmp_device_ref) reflecting what the functions actually access.
- Remove @dialyzer :nowarn_function on three agent_channel helpers.
remote_ip.ex — real bug caught and fixed:
- `:ranch.get_addr(socket.transport_pid)` was always raising since
Bandit uses ThousandIsland, not Ranch; the rescue _ -> nil silently
returned nil every time. Switched to Phoenix's documented
:peer_data connect_info (already enabled in endpoint.ex) via
socket.assigns; remote IP now actually works.
- Remove remote_ip.ex entry from .dialyzer_ignore.exs.
Accounts / Organizations (Ecto.Multi opacity):
- Add @specs to Multi-building helpers, refactor into pipe chains.
- 6 @dialyzer :nowarn_function → 0, but 7 :no_opaque remain. Root
cause is upstream: Ecto.Multi.new/0 returns a struct with a literal
%MapSet{} whose @opaque internal representation trips dialyzer on
every subsequent Multi.* call. Unfixable without an Ecto patch or
bypassing Multi entirely. Comments document the specific upstream
issue rather than a vague "Ecto.Multi opacity" claim.
Devices.ex:
- Adding @specs made it worse (call_without_opaque → contract_with_
opaque); inlining the Multi didn't help either — same MapSet root
cause. Suppression kept with a sharper comment.
Categories addressed:
- pattern_match / pattern_match_cov (32): remove dead case/with clauses
that dialyzer proved unreachable from the caller types.
- contract_supertype / extra_range / invalid_contract /
contract_with_opaque (25): narrow @spec declarations to match actual
success typings.
- call / call_without_opaque (18): fix bad calls, narrow User.t to
allow nil for in-memory changeset structs, suppress Ecto.Multi
opaque-type false positives with targeted @dialyzer directives.
- guard_fail / no_return / unused_fun / unknown_function (13): remove
dead || fallbacks, simplify always-true params, cascade-resolve
no_returns via the underlying pattern_match and call fixes.
Real production bug fixed: StormDetector.handle_cast/2 had swapped
`:queue.in` args (`queue |> :queue.in(ts)` which desugars to
`:queue.in(queue, ts)` — wrong argument order). Alert timestamps
were never being enqueued, so storm detection would fail at runtime.
Corrected to `ts |> :queue.in(queue)`.
.dialyzer_ignore.exs: suppress two genuine dep-PLT gaps
(:ranch.get_addr/1 false positive from Bandit's transitive ranch,
and the Cloak.Vault GenServer callback_info on the CI build path).
`mix dialyzer` now: Total errors: 114, Skipped: 114 — passes clean.
Warnings: 88 → 0.
Prefix fire-and-forget calls (PubSub.broadcast, Oban enqueue, Logger,
update_*, etc.) with `_ =` so dialyzer knows the return value is
intentionally discarded. Wrap a few if/case expressions whose
branches produce mixed types the same way.
No behavior changes — only explicit acknowledgement of discarded
returns.
Warnings: 242 → 88.
Phase 1: Foundation Types (100% complete)
- query_helpers: SQL LIKE sanitization with pipe operators
- numeric: Integer parsing with pattern matching guards
- result: Pure Elixir Result monad (map, and_then, unwrap_or)
Phase 2: Ecto Domain Types (100% complete)
- ip_address: IPv4/IPv6 validation using :inet directly
- mac_address: Multi-format MAC parsing (colon/hyphen/dot/compact)
- snmp_oid: OID parsing/manipulation with recursive pattern matching
All 198 tests passing across converted modules.
API changed from Gleam-style {:some/:none to idiomatic {:ok/:error.
Refactored parse_numeric_oid to use with statement, reducing nesting depth.
Reviewed-on: graham/towerops-web#196
- Check device.snmp_enabled before creating discovery/polling jobs
- Also check snmp_enabled for MikroTik jobs
- Prevents agent from attempting SNMP operations on ping-only devices
- Fixes "Agent collected 0 OIDs" errors for devices without SNMP
Reviewed-on: graham/towerops-web#180
Devices configured as SNMPv1 were being force-upgraded to v2c for agent
polling, which breaks devices that don't support v2c (returning empty
OID responses and no throughput/capacity data).
Reviewed-on: graham/towerops-web#166
When an agent joins, we update last_seen_at in the database but don't set
last_heartbeat_db_update in the socket. This causes the heartbeat throttling
logic to be out of sync:
- On join: DB updated, but last_heartbeat_db_update not set (nil)
- First heartbeat: Updates DB again because last_heartbeat_db_update is nil
- Subsequent heartbeats: Check if >30s since last DB update
- Problem: last_heartbeat_db_update tracks the first heartbeat, not the join DB update
This manifests when Phoenix recompiles during development - the channel process
continues running with old socket assigns, and if last_heartbeat_db_update is
recent, heartbeats won't update the DB. After 10+ minutes without DB updates,
the agent appears offline even though it's still connected.
Fix: Set last_heartbeat_db_update when we update the DB on join, so the
throttling logic accurately tracks the most recent DB update.
Reviewed-on: graham/towerops-web#142
The Check protobuf has a `oneof config` field which generates separate
fields in the Elixir struct: :http, :tcp, :dns, :ssl. However,
check_type_config/2 was returning [config: {:http, ...}] which doesn't
match the struct definition, causing a KeyError crash in
build_and_push_check_jobs/1.
This bug was originally fixed in e0a74e2f but was accidentally
reintroduced in f703e61b when attempting to use "tagged tuple config".
Fix: Return field-specific keyword lists (e.g., [http: %HttpCheckConfig{}])
instead of [config: {:http, ...}].
Also added comprehensive test that creates all 4 check types and verifies
no KeyError crash occurs when sending check jobs to agent.
Reviewed-on: graham/towerops-web#141
check_type_config was returning `config: {:http, ...}` but the
Check struct has no :config field — it has :http, :tcp, :dns, :ssl
directly. This caused a KeyError crash in build_and_push_check_jobs.
Reviewed-on: graham/towerops-web#121
Replace the protobuf hex package (agent.pb.ex) and hand-written validator
(validator.ex) with a pure Gleam implementation: wire format primitives,
all 23 message types, encode/decode functions, and validation merged into
decode. Thin Elixir wrappers in agent.ex preserve the existing struct API.
- Wire format: varint, length-delimited, double (IEEE 754 via FFI), tag encode/decode
- Types: all proto3 message types as Gleam custom types with enums and oneof
- Decode: field-level parsing with integrated validation (UUID, IP, hostname, limits)
- Encode: proto3 default omission, bytes_tree concatenation, map/repeated fields
- Wrappers: Elixir structs with encode/decode delegating to Gleam, enum atom conversion
- Remove {:protobuf, "~> 0.12"} dependency
Reviewed-on: graham/towerops-web#104
Set up mix_gleam archive, gleam_stdlib, and gleeunit deps. Add Gleam
compiler to all Dockerfiles, CI workflows, nix shell, and .tool-versions.
Rewrite Numeric, QueryHelpers, and URLValidator from Elixir to Gleam with
full ExUnit test coverage. Update all callers to use the Gleam modules
directly via :towerops@module syntax. Remove duplicate sanitize_like/1
private functions in snmp.ex and trace.ex.
Reviewed-on: graham/towerops-web#98
Protobuf oneof fields use a single :config key with tagged tuples
like {:dns, %DnsCheckConfig{}} rather than individual keys. The
check_type_config functions were setting e.g. dns: %DnsCheckConfig{}
directly, causing KeyError crashes in Check.__struct__/1.
Reviewed-on: graham/towerops-web#101
add ssl check executor that connects via TLS, reads peer certificate
expiry, and returns OK/WARNING/CRITICAL based on configurable
warning_days threshold. includes protobuf config, liveview form fields,
agent channel encoding, oban worker dispatch, and unit tests.
Reviewed-on: graham/towerops-web#64
send check_jobs protobuf to agents during job dispatch cycle, handle
check_result messages back with state transitions and alert
creation/resolution. auto-assign checks to device's effective agent
token on creation, broadcast check changes via PubSub so connected
agents get immediate updates.
add edit/delete support for service checks on device show page with
form pre-population from existing check config. auto-fill TCP host
from device IP address.
includes validate_check_result/1 for protobuf validation and
list_checks_for_agent/2 for querying agent-assigned service checks.
also fixes pre-existing broken device_type_icon reference in device
index template.
Reviewed-on: graham/towerops-web#50
- add MobileSocket for token-authenticated mobile WebSocket connections
- add MobileChannel with org-level and device-level real-time events
- add /mobile/socket endpoint for iOS app connections
- add "Link Mobile App" to user dropdown menu for discoverability
- improve QR code page with numbered steps and clearer instructions
- add socket/channel tests (17 total)
The server only sent polling/ping jobs to agents once on WebSocket join
and never re-dispatched them. After the first batch completed (~5-10s),
devices were never polled again until the agent reconnected.
:send_jobs now schedules a follow-up dispatch every 60s (configurable).
Unifies the timer management with :assignments_changed to prevent
accumulation.
Previous fix only resolved alerts when device status changed from down to up.
This didn't help devices already marked as 'up' with stuck unresolved alerts.
Now:
- Always check for stuck device_down alerts after successful SNMP poll
- Also check for stuck device_up alerts (should always be auto-resolved)
- Logs when stuck alerts are found and resolved
- Works for existing stuck alerts without waiting for status to change
This fixes the 8 stuck alerts in production - they'll auto-resolve on next poll.
When agents poll devices via SNMP and get successful results, the code was
only processing sensor/interface data without updating device status or
resolving alerts. This caused device_down alerts to persist even after
devices came back online.
This fix:
- Updates device status to 'up' when SNMP polling succeeds
- Auto-resolves any active device_down alerts
- Creates device_up alert (matching DeviceMonitorWorker behavior)
- Broadcasts device_status_changed via PubSub
Also fixes device_up alert creation to include resolved_at timestamp
(matching DeviceMonitorWorker behavior).
Test coverage: Added test verifying auto-resolution when device recovers.
CRITICAL fixes:
- Mask SNMP community string in agent channel logs (CRITICAL-1)
- Remove snmpv3 passwords from REST API responses, return _set booleans (CRITICAL-2)
- Replace snmp_community with snmp_community_set in GraphQL type (CRITICAL-3)
HIGH fixes:
- Fix cookie same_site from invalid 'Towerops' to 'Lax' (HIGH-4)
- Remove unsafe-eval from CSP script-src (HIGH-6)
- Block GraphQL introspection queries in production (HIGH-7)
- Sanitize LIKE wildcards in SNMP device name search (HIGH-8)
- Reject webhooks when no secret configured instead of accepting (HIGH-9)
MEDIUM fixes:
- Hash mobile session tokens (SHA-256) before DB storage (MEDIUM-10)
- Apply security headers in all environments, not just prod (MEDIUM-14)
- Add GraphQL query complexity limit (500) in production (MEDIUM-16)
- Fix X-Frame-Options to DENY to match frame-ancestors 'none' (MEDIUM-13)
- Fix broken route paths in dashboard (/discovery, /subscribers/trace, /config-changes)
- Fix insights empty state settings link (org-scoped route)
- Add underscores to all 86400 literals across 6 files
- Alphabetize aliases in search.ex and agent_channel.ex
- Extract changelog parser helper to reduce nesting
- Extract dashboard impact calculation to reduce nesting
- Refactor agent_channel config change detection (pattern match, extract function)
- Combine double Enum.reject into single call in trace.ex
- Fix line length in trace.ex search query
- Replace length/1 > 0 with != [] in trace_live
- Run mix format on all files
- create MonitoringCheck schema for the monitoring_checks table, fixing
silent data loss where ping results were dropped by the wrong schema
- update agent channel and device monitor worker to use new schema
- fix Stats queries to use MonitoringCheck instead of Check schema
- add list_online_cloud_pollers and list_cloud_polled_devices queries
- add CloudLatencyProbeWorker to dispatch cross-agent latency probes
every 8 hours via PubSub to all online cloud pollers
- scope AgentLatencyEvaluator to cloud pollers only with cloud_only
filter, lower min_checks (5), wider time window (48h), and
device-level assignments to avoid changing site/org defaults
- update cron schedule: probes at 0/8/16h, evaluation at 0:30/8:30/16:30
- refactor monitoring.ex and http_executor.ex to fix credo complexity
Enhanced AgentChannel logging to track device IDs across job list updates.
Added tests verifying server correctly excludes deleted devices from job lists.
Root cause: Bug is in Go agent (appends jobs without clearing), not Phoenix server.
Always broadcast from update_device_status/2 so the device index
LiveView refreshes even when status hasn't changed. Add a 15-second
tick timer to keep "Last Checked" timestamps fresh between polls.
Scope alert PubSub topics to organization for defense-in-depth.
Replace individual Repo.insert() with Repo.insert_all() for sensor readings,
interface stats, processor readings, and storage readings in agent channel and
device poller worker. Add partial/composite database indexes for common query
patterns. Dashboard now uses GROUP BY aggregation instead of loading all devices.
DeviceLive.Show only reloads data for the active tab on PubSub events.
DeviceLive.Index debounces status changes and skips quota queries on updates.
Agent polling preloads filtered to monitored sensors/interfaces only.
- use 64-bit HC counters (ifHCInOctets/ifHCOutOctets) instead of 32-bit
counters that wrap every ~34s on gigabit links
- fix negative integer parsing in replay adapter so sensor scale values
like -3 produce correct divisors (1000) instead of falling through to 1
- replace walking entire Cisco enterprise tree with targeted subtrees to
avoid timeouts; add Juniper, HP/H3C, Fortinet, Cambium vendor OIDs
- add missing discovery OIDs: ENTITY-STATE-MIB, HOST-RESOURCES-MIB
tables, and UCD-SNMP-MIB Linux CPU stats fallback
strip leading dots from OID keys returned by agent so they match
sensor OIDs stored without leading dots. fixes ubiquiti airos
wireless sensors (clients, ccq, frequency, etc) not being polled.
also fix flaky sensor_change_detector tests using global pubsub topic.
extract sensor change detection from DevicePollerWorker into shared
SensorChangeDetector module. add sensor_value_changed event for non-%
sensors when value differs from last_value. agent channel now detects
changes and updates last_value, matching phoenix poller behavior.
division by sensor_divisor produced full float precision values
like 34.09999999. apply Float.round/2 in both the phoenix poller
and agent channel paths.
store_monitoring_check saved ping results from agents but never
updated device status. DeviceMonitorWorker skips agent-assigned
devices, so nothing ever moved them from :unknown to :up/:down.
added update_device_status_from_check to derive status from check
result, update the device, create alerts on transitions, and
broadcast via PubSub — matching DeviceMonitorWorker behavior.
newly discovered devices sat idle with no metric data until the
next scheduling cycle. now the channel sends a poll job (and ping
job if monitoring enabled) right after discovery succeeds, using
send(self(), ...) so discovery DB writes commit first.
also reduces k8s replicas from 3 to 2.
