Root cause: AF11X devices only support SNMPv1, IF-MIB eth0 counters
return zero, and air0 32-bit counters wrap too fast for backhaul speeds.
Fixes:
- Use correct UBNT-AirFIBER-MIB OID indices from MIB SEQUENCE definition:
.7.1 = rxOctetsOK (inbound), .6.1 = txOctetsOK (outbound)
.10.1 = rxErroredFrames, .11.1 = txErroredFrames
- Detect AirFiber devices by sysDescr instead of SNMP probing (v1 devices
cannot return Counter64 via standard probe queries)
- Force SNMPv1 for AirFiber proprietary queries
- Apply proprietary counters to all real interfaces (not just eth0)
Verified against live AF11X at 10.250.1.90:
- IF-MIB eth0 returns zero counters
- Proprietary .6.1/.7.1 show real traffic (3.75TB rx / 43TB tx)
Reviewed-on: graham/towerops-web#170
- Clamp negative counter deltas to zero instead of attempting 32-bit wrap
correction (matches LibreNMS behavior — wraps are indistinguishable from
HC counter resets, so discard the ambiguous data point)
- Add rate sanity check: discard values exceeding 400 Gbps as counter anomalies
- Add 4-byte binary decoding for 32-bit SNMP counters in decode_snmp_value
- Track HC vs standard counter usage per interface stat (is_hc field)
- Add migration for is_hc column on snmp_interface_stats
- Update tests to match new clamping behavior
Reviewed-on: graham/towerops-web#168
The wrap detection only applied the 2^32 correction when the previous
counter was in the upper half of the 32-bit range (> 2^31). On fast
links, counters wrap regardless of where in the range they sit. Three
graph rendering paths had no wrap detection at all, just max(0).
Consolidated all bps calculations into a single public
Capacity.calculate_bps/3 that always tries the 2^32 correction for
negative deltas. 64-bit HC counter resets remain clamped to zero since
the corrected delta stays negative.
Reviewed-on: graham/towerops-web#163
Standard ifInOctets/ifOutOctets are 32-bit counters that wrap around
4,294,967,296 bytes (~4.3 GB). At 1 Gbps this wraps every ~34 seconds,
causing brief drops to 0 bps in traffic charts and capacity reports.
When prev > 2^31 and delta is negative, add 2^32 to recover the correct
rate. If the result is still negative (HC 64-bit counter after reboot),
clamp to 0. Small-value decreases (prev < 2^31) continue to clamp to 0
since they cannot be real 32-bit wraps.
Fixes both capacity.ex (utilization reports) and device_live/show.ex
(per-device traffic chart).
Reviewed-on: graham/towerops-web#161
Add data_retention_days field to organizations (default 365, range 30-730).
DataRetentionWorker runs nightly at 1 AM to purge expired time-series data
across 8 tables, respecting each org's retention setting.
TimescaleDB global retention bumped from 90 to 730 days as safety net.
Resolved alerts are cleaned up; unresolved alerts kept regardless of age.
Reviewed-on: graham/towerops-web#144
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
CRITICAL FIXES:
- Fix unsafe pattern matching in SNMP discovery that would crash on timeout
- Changed {:ok, _} = ... to proper case statements with error handling
- Extracted safe_discover/3 helper to reduce cyclomatic complexity
- Added fallback to empty lists and warning logging for timeouts
- Affects: discover_vlans, discover_ip_addresses, discover_processors, discover_storage
MEMORY LEAK FIXES (JS Hooks):
- GlobalSearchTrigger: Added destroyed() cleanup for click listener
- SidebarCollapse: Added destroyed() cleanup for click listener
- ThemeSelector: Added destroyed() cleanup for phx:set-theme window listener
- NetworkMap: Added destroyed() cleanup for 3 button listeners (zoom in/out/fit)
- All hooks now properly store handler references and remove listeners on unmount
LIVEVIEW FIXES:
- Added phx-update="ignore" to 5 SensorChart hooks to prevent chart re-initialization
- Added catch-all handle_info(_msg, socket) to 5 LiveViews to prevent crashes on unexpected messages
- device_live/show.ex, device_live/index.ex, alert_live/index.ex, activity_feed_live.ex, agent_live/index.ex
N+1 QUERY FIXES:
- Created Gaiia.get_site_subscriber_summaries/1 batch function
- Refactored alert_live and device_live to use batch query instead of looping
- Reduces queries from N to 1 when loading site subscriber data
Impact:
- Prevents discovery worker crashes during SNMP timeouts
- Eliminates memory accumulation from leaked event handlers
- Prevents chart state loss on LiveView updates
- Prevents LiveView crashes from unexpected PubSub messages
- Improves database performance by eliminating N+1 queries in alert/device views
- Passes credo --strict with 0 issues
Reviewed-on: graham/towerops-web#132
- Wrap upsert_nodes in Repo.transaction so all upserts share one DB
connection, preventing pool exhaustion under load
- Remove stale device_role_source reference from form test
Reviewed-on: graham/towerops-web#118
- Reduce device type options from 10 to 6 (server, switch, router, access_point, backhaul, other)
- Change label from "Device Role" to "Device Type"
- Disable auto-detection - device type is now required and manual-only
- Add migration to map existing device roles to new simplified set
- Update REST API to include device_role and device_role_source fields
- Automatically set device_role_source to "manual" when device_role is provided
- Update wireless detection to include "backhaul" type
- Enhance display formatting for "Access Point"
Breaking Changes:
- Auto-inference of device type has been disabled
- device_role is now a required field in forms
- GraphQL/REST API consumers will see changed device_role values after migration
Files Changed:
- lib/towerops_web/live/device_live/form.html.heex
- lib/towerops/topology.ex
- lib/towerops/snmp/discovery.ex
- lib/towerops/devices/device.ex
- lib/towerops_web/live/device_live/index.ex
- lib/towerops_web/controllers/api/v1/devices_controller.ex
- priv/repo/migrations/20260322152809_migrate_device_roles_to_simplified_set.exs
- test/towerops/topology_test.exs
- test/towerops_web/live/device_live/form_test.exs
Reviewed-on: graham/towerops-web#109
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
Migrate three pure-function modules from Elixir to Gleam:
- polling_offset: deterministic hash-based polling offset (pure Gleam, no FFI)
- changelog_parser: text parsing logic in Gleam, thin Elixir wrapper for File I/O
- user_agent_parser: regex-based UA parsing in Gleam with Erlang FFI for
atom-keyed map conversion
Add gleam_regexp dependency for regex support in Gleam modules.
Update all call sites and tests to use Gleam module atoms.
Reviewed-on: graham/towerops-web#103
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
## Summary
- **DB pool exhaustion**: `device_has_effective_agent?` was doing 2 DB queries per call (3-table join + global settings lookup), called 2-3x per worker cycle across hundreds of devices (~6000 unnecessary queries/minute). Added an ETS-backed `AgentCache` with 120s TTL and removed redundant calls within the same worker cycle.
- **Settings page crash**: `KeyError: key :user_agent not found` on BrowserSession — field doesn't exist, replaced with `device_type`.
- **Login history IP truncation**: Changed from side-by-side grid to stacked table layout so IP addresses aren't clipped.
## Test plan
- [x] 8 new tests for AgentCache (cache hit, miss, invalidation, global default)
- [x] Full test suite passes (8651 tests, 0 failures, verified across multiple runs)
- [x] `mix format` and pre-commit hooks pass
Reviewed-on: graham/towerops-web#88
- Wrap agent name + icon inside the offline amber badge so it's clear
what is offline on the device detail page
- Add "Agent:" label before the agent indicator for all states
- Fix TimeSeriesTest deadlock: switch to async: false since check_results
is a TimescaleDB hypertable and concurrent chunk creation deadlocks
- Fix unicode property test: String.length/1 counts grapheme clusters,
not code points; combining chars can make a 2-codepoint string have
only 1 grapheme, legitimately failing the min: 2 validation
Reviewed-on: graham/towerops-web#83
## Summary
- **`list_checks_for_agent` cascade**: the agent query only matched checks with `agent_token_id` set directly on the check record — checks on devices inheriting their agent via site/org/global default were never sent to the Go agent. Now uses left joins with the full cascade (device assignment → site → org default → global default cloud poller), matching the pattern from `list_agent_polling_targets`.
- **At-risk summary double-counting**: `get_impact_summary` called `analyze_device_impact` per down AP, and each returned the full site subscriber count. With 3 APs down at a 147-sub site it showed 441 subs / $31k instead of 147 / $10k. Fixed by deduplicating impact per site. Same fix applied to `calculate_down_impact` in site impact summaries.
## Test plan
- [x] 3 new cascade tests for `list_checks_for_agent` (site, org default, global default)
- [x] 3 new tests for `get_impact_summary` (single site dedup, multi-site sum, zero case)
- [x] All 8,607 existing tests pass
- [ ] Deploy to staging, verify At Risk numbers on dashboard
- [ ] Verify Go agent receives DNS/HTTP checks via cascade
Reviewed-on: graham/towerops-web#79
## Summary
- Add `device_has_effective_agent?/1` to Agents context that checks full cascade (device → site → org → global default) without distinguishing between cloud pollers and local agents
- Update check_executor_worker, device_poller_worker, and device_monitor_worker to use new function
- Phoenix never executes checks (DNS, ping, HTTP, TCP, SSL, SNMP) when any agent handles the device
## Problem
The global default cloud poller was not being checked in `should_phoenix_poll_device?/1`, causing Phoenix to run DNS/ping/service checks server-side even when the Go agent was handling the device. This produced duplicate results with NXDOMAIN errors from the Erlang DNS resolver.
## Test plan
- [x] 6 new tests for `device_has_effective_agent?/1` covering all cascade levels
- [x] Full test suite passes (8601 tests)
- [ ] Verify staging DNS check stops showing NXDOMAIN errors after deploy
Reviewed-on: graham/towerops-web#77
Check.changeset/2 did not include state fields (current_state,
current_state_type, current_check_attempt, last_check_at, etc.) in
its cast list. Ecto silently dropped all state updates, so
update_check_state succeeded but wrote nothing to the database.
Added Check.state_changeset/2 for internal state machine updates,
keeping the regular changeset safe from user input. Updated
Monitoring.update_check_state/3 to use it directly.
Reviewed-on: graham/towerops-web#73
## Summary
- **DNS checks**: `parse_server` returned a flat 5-tuple `{a,b,c,d,53}` instead of `{{a,b,c,d}, 53}` — the format `:inet_res` requires for nameservers. All DNS checks using a custom server failed with argument error.
- **Ping checks**: `System.cmd` `:timeout` option not supported on Elixir 1.19.4 (staging). Replaced with `Task.async`/`Task.yield` wrapper for cross-version compatibility.
- **Checks table UI**: Removed unnecessary flex div wrapper on status badge column that caused vertical misalignment with other table cells.
## Test plan
- [x] DNS executor tests pass (custom server resolution now succeeds)
- [x] Ping executor tests pass (no more ArgumentError)
- [x] Full test suite passes (8593 tests, 0 new failures)
Reviewed-on: graham/towerops-web#72
## Summary
- **Ping checks never scheduled**: `ensure_default_ping_check` created check records but never called `schedule_check`, so auto-created ping checks never executed
- **Skipped checks permanently orphaned**: `CheckExecutorWorker` only rescheduled checks that actually executed — skipped checks (agent-assigned, SNMP disabled) broke the scheduling chain permanently
- **No recovery mechanism**: `JobHealthCheckWorker` only recovered `DeviceMonitorWorker` jobs, not `CheckExecutorWorker` jobs — any lost check job was gone forever
## Changes
- `monitoring.ex`: Call `schedule_check` after creating ping check in `ensure_default_ping_check`
- `check_executor_worker.ex`: Reschedule skipped-but-enabled checks so conditions are re-evaluated next cycle
- `job_health_check_worker.ex`: Recover orphaned `CheckExecutorWorker` jobs for enabled checks (runs every 10 min)
- `devices.ex`: Wrap `ensure_default_ping_check` in try/rescue for deadlock resilience
## Test plan
- [x] New tests: ping check scheduling on creation, no duplicate scheduling on idempotent call
- [x] Updated tests: skipped checks now assert rescheduling instead of orphaning
- [x] New tests: health check worker recovers orphaned check executor jobs, skips disabled checks
- [x] All 177 targeted tests pass
- [x] Compile clean with warnings-as-errors
- [x] Pre-commit hooks pass (credo, format)
Reviewed-on: graham/towerops-web#71
## Summary
- Every device now automatically gets an ICMP ping check (`source_type: "auto_discovery"`) when created via `Monitoring.ensure_default_ping_check/1`
- When a device IP changes, the ping check host config is synced automatically
- Backfill migration inserts ICMP Ping checks for all existing devices that don't have one
- Fixes checkbox alignment in the Add Service Check modal (verify SSL / follow redirects)
## Test plan
- [x] 4 new tests for `ensure_default_ping_check/1` (create, idempotent, IP update, no-op)
- [x] 3 new tests for device lifecycle (auto-create on device create, sync on IP change, no-op when IP unchanged)
- [x] 13 existing tests updated to account for auto-created ping check
- [x] Full test suite passes (`mix precommit`)
- [ ] Verify backfill migration on staging
- [ ] Create device in UI, confirm ping check appears in service checks list
- [ ] Verify checkbox alignment in Add Service Check modal
Reviewed-on: graham/towerops-web#68
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
- Skip HTTP/TCP/DNS checks in Phoenix when device is assigned to agent
- Agents now execute service checks from the device's network location
- Fix cancel button in service check modal (text link instead of button)
- Fix status badge alignment in checks table
Reviewed-on: graham/towerops-web#63
add early bail in DiscoveryWorker when snmp_enabled is false.
mark snmp_not_enabled and device_unresponsive as non-retryable
errors to prevent wasted retries.
Reviewed-on: graham/towerops-web#58
- REST CRUD at /api/v1/checks for HTTP, TCP, DNS, ping checks
- GraphQL queries (checks, check) and mutations (create/update/delete)
- ping executor using system ping with macOS/Linux parsing
- check config validation for ping type (requires host)
- API documentation updated with checks resource section
Reviewed-on: graham/towerops-web#52
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
Extend the GraphQL API and socket to accept mobile session tokens
alongside existing API tokens. Add OrganizationScope middleware
that extracts organization_id from query args for mobile users.
Add my_organizations query for mobile app org listing.
Reviewed-on: graham/towerops-web#48
the complete_qr_login endpoint was returning session.token (the SHA256
hash stored in the database) instead of session.raw_token (the plaintext).
this caused all subsequent API calls and WebSocket connections to fail
with 401/REFUSED since the server would hash the already-hashed token.
adds a regression test that verifies the returned token is usable for
authenticated API requests.
Reviewed-on: graham/towerops-web#42
- Changed sidebar from fixed to sticky positioning
- Wrapped sidebar and main content in flex container
- Sidebar now sticks to top along with header when scrolling
- Removed unused Ecto.Query import from gps_sync.ex
Credo improvements (17 → 0 remaining):
- Replaced all length/1 checks with empty list comparisons
- Extracted helper functions to reduce nesting depth
- Split complex functions into smaller, testable units
- Applied 'with' statements for clearer control flow
- Refactored high-complexity functions (cyclomatic complexity)
- Files improved: storm_detector, alert_notification_worker,
alert_digest_worker, gps_sync, statistics_sync, device_sync,
site_sync, site_correlation, reports_live, topology,
pagerduty/client, cn_maestro/sync
Reviewed-on: graham/towerops-web#32
The StormDetector was being added twice:
1. In the main children list (line 127)
2. In background_workers() (line 176)
This caused the supervisor to fail with 'duplicate_child_name' error.
Removed the duplicate from main children list since it's correctly
placed in background_workers() which also handles test env properly.
- StormDetector GenServer: sliding-window per org, >10 alerts/min triggers
suppression mode with single summary notification
- Site correlation: >3 devices at same site down within 120s creates
single 'Site Outage' alert instead of N individual alerts
- Notification rate limiter: max 5 per user per 15min window, excess
batched into digest via AlertDigestWorker
- PagerDuty client: retry with exponential backoff on 429, respects
Retry-After header
- Batch maintenance checks: new Maintenance.devices_in_maintenance/1
takes list of device_ids, returns MapSet (single query vs N queries)
- Migration: alert_storms, site_outages, notification_digests tables
plus site_outage_id/storm_suppressed on alerts
- DeviceMonitorWorker integrates storm detection and site correlation
before creating alerts
- Tests for StormDetector, SiteCorrelation, NotificationRateLimiter,
and batch maintenance checks
