Skip to content
django-micboarddjango-micboard

SRED Project Summary — 2026 Bound Live Monitoring Projections

Project Description

Charger dashboards and display-wall kiosks refreshed repeatedly through HTMX. Loading every charger, slot, section, unit, and performer made memory and response size proportional to tenant inventory. Tenant filtering prevented data disclosure but did not bound work inside a large authorized tenant. Query counts scaled with data volume, causing timeouts on large deployments.

Project Goals

Build live pages from primitive Pydantic snapshot DTOs rather than passing ORM graphs to templates. Apply tenant visibility before every cutoff and use stable business ordering with primary-key tie breakers. Enforce package ceilings: charger dashboard (64 chargers, 32 slots/charger); display wall (16 sections, 32 chargers/section, 32 occupied slots/charger); display-wall health (16 sections, 32 chargers/section, 32 slots/charger). Fetch one sentinel row beyond each ceiling so snapshots report truncation without counting the full relation. Render accessible overflow notices from DTO truncation metadata. Resolve performers/units only for identities present in the bounded visible window; when multiple active assignments exist, rank in SQL and materialize only the winner per unit ordered by priority, update time, then primary key.

Technical Uncertainties

Uncertainty #1: Sentinel-Based Truncation Detection Without Full Count

Description: Needed to report “X more items” in UI without executing COUNT(*) on the full relation (defeats the bounding purpose). The challenge: detect truncation while fetching only ceiling+1 rows.

Experiments:

  • Attempted: window function ROW_NUMBER() with WHERE rn <= ceiling + 1 — worked but added query complexity
  • Adopted: queryset[:ceiling+1] + Python-side len(results) > ceiling → truncated = True, visible = results[:ceiling]; ORM LIMIT ceiling+1 is a single query

Results / Learnings / Success:

  • Constant 1 query per projection regardless of tenant size
  • Truncation metadata (truncated=True, total_estimate="100+") in DTO for accessible notice
  • Query-budget tests enforce constant query count for empty and 10K+ item tenants

Uncertainty-Specific Documentation & Links:

Uncertainty #2: Performer Assignment Resolution Within Bounded Window

Description: Multiple active PerformerAssignment records can exist for one WirelessUnit (handoffs, overlaps). In unbounded view, all were fetched. In bounded view, only assignments for units in the visible window should be resolved — but the “winning” assignment must be ranked globally, not just within the window.

Experiments:

  • Subquery: ROW_NUMBER() OVER (PARTITION BY unit_id ORDER BY priority DESC, updated_at DESC, pk) in assignment CTE; join to units in visible window; filter rn = 1
  • Single query returns units + their top-ranked performer; no post-processing

Results / Learnings / Success:

  • Assignment resolution scales with window size (32), not inventory size (10,000)
  • Global ranking preserved; window only limits materialization

Uncertainty-Specific Documentation & Links:

Participants

Name Role % Yearly Time Contribution
(team lead) Architecture / DTO Design ~25% Ceiling design, sentinel pattern, ADR-012
(engineer) Implementation / Query Budgets ~35% DTOs, projections, query-budget tests
(engineer) Implementation / Overflow UI ~25% Accessible truncation notices, HTMX fragments

Project Documentation & Links

Project Docs:

PRs:

  • (projection DTOs PR)
  • (charger dashboard bounding PR)
  • (display wall bounding PR)
  • (query budget tests PR)
  • (overflow UI PR)