Fix remaining get_client_ip stale imports; split Phase 5 to work_queues.md

Import fixes (all three imported get_client_ip/ratelimit_ip from sapl.utils
which no longer exports them — causing the ImportError at startup):
- sapl/materia/forms.py: move get_client_ip to sapl.middleware.ratelimit
- sapl/materia/views.py: move get_client_ip + ratelimit_ip; keep
  RATE_LIMITER_RATE in sapl.settings (used by @ratelimit decorators)
- sapl/base/views.py: same pattern as materia/views.py

Docs:
- rate-limiter-v2.md: remove Phase 5 section (§8); renumber Open Questions
  to §8; update Table of Contents
- work_queues.md (new): Async PDF via Celery + Django Channels WebSocket
  voting panel, with full context, Redis B topology rationale, k8s manifest
  list, and open questions. Planned start: after rate-limiter-2026 is stable.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

rate-limiter-v2.md

@@ -16,8 +16,7 @@
 6. [Phase 2 — Rate Limiting & Bot Mitigation](#5-phase-2--rate-limiting--bot-mitigation)
 7. [Phase 3 — File Serving Corrections](#6-phase-3--file-serving-corrections)
 8. [Phase 4 — Dynamic Page Caching](#7-phase-4--dynamic-page-caching)
-9. [Phase 5 — Async PDF & WebSocket (Follow-up)](#8-phase-5--async-pdf--websocket-follow-up)
+9. [Open Questions](#8-open-questions)
-10. [Open Questions](#9-open-questions)
 
 ---
 
@@ -1131,91 +1130,7 @@ class PesquisarMateriaView(FilterView):
 
 ---
 
-## 8. Phase 5 — Async PDF & WebSocket (Follow-up)
+## 8. Open Questions
-
-**Goal**: Eliminate synchronous PDF generation as a memory pressure source; add WebSocket support.  
-**Prerequisite**: Phase 1 (Redis deployed). WebSocket work resumes **after** Redis is on k8s, bot siege is resolved, and OOM pressure is reduced.
-
-### 8.1 Async PDF via Celery
-
-Current synchronous flow — holds worker memory for entire PDF build:
-
-```mermaid
-sequenceDiagram
-    participant B as Browser
-    participant G as Gunicorn worker
-    participant ORM as PostgreSQL
-    participant RL as ReportLab
-
-    B->>G: GET /pdf/materia/12345
-    G->>ORM: N+1 queries (get_etiqueta_protocolos)
-    ORM-->>G: data
-    G->>RL: build entire PDF in RAM
-    RL-->>G: PDF bytes (held in worker memory)
-    G-->>B: stream response
-    note over G: worker blocked for full duration
-```
-
-Target async flow — worker freed immediately:
-
-```mermaid
-sequenceDiagram
-    participant B as Browser
-    participant G as Gunicorn worker
-    participant Q as Redis (Celery queue)
-    participant W as Celery worker
-    participant D as Disk
-
-    B->>G: POST /pdf/materia/12345
-    G->>Q: enqueue task
-    G-->>B: 202 Accepted + task_id
-    W->>W: build PDF (out of band)
-    W->>D: write PDF to disk
-    B->>G: GET /pdf/status/task_id
-    G-->>B: 302 → nginx /media/pdf/task_id.pdf
-```
-
-### 8.2 Celery Configuration
-
-> **Critical**: Celery broker **must** be a **separate** Redis instance (Redis B) with `noeviction` policy. The cache Redis (Redis A) uses `allkeys-lru` — tasks would silently disappear if evicted under memory pressure.
-
-```yaml
-# docker/k8s/redis-celery-configmap.yaml
-data:
-  redis.conf: |
-    maxmemory-policy noeviction   # never evict tasks
-    appendonly yes                 # AOF persistence ON
-    save "900 1"                   # RDB snapshot
-```
-
-```python
-# sapl/settings.py
-CELERY_BROKER_URL    = config('CELERY_BROKER_URL',    default='')
-CELERY_RESULT_BACKEND = config('CELERY_RESULT_BACKEND', default='')
-```
-
-### 8.3 Django Channels (WebSocket Voting Panel)
-
-Uses Redis DB2 on the same Redis A instance (cache + rate limiter pod):
-
-```python
-# sapl/settings.py
-CHANNEL_LAYERS = {
-    "default": {
-        "BACKEND": "channels_redis.core.RedisChannelLayer",
-        "CONFIG": {
-            "hosts": [("sapl-redis.redis.svc.cluster.local", 6379)],
-            "db": 2,      # DB2 reserved for channels
-            "capacity": 1500,
-            "expiry": 10,
-        },
-    }
-}
-```
-
----
-
-## 9. Open Questions
 
 | # | Question | Status | Blocks |
 |---|---|---|---|

sapl/base/views.py

@@ -51,7 +51,8 @@ from sapl.sessao.models import (Bancada, SessaoPlenaria)
 from sapl.settings import EMAIL_SEND_USER, RATE_LIMITER_RATE
 from sapl.utils import (gerar_hash_arquivo, intervalos_tem_intersecao, mail_service_configured,
                         SEPARADOR_HASH_PROPOSICAO, show_results_filter_set, google_recaptcha_configured,
-                        get_client_ip, sapn_is_enabled, is_weak_password, ratelimit_ip)
+                        sapn_is_enabled, is_weak_password)
+from sapl.middleware.ratelimit import get_client_ip, ratelimit_ip
 from .forms import (AlterarSenhaForm, CasaLegislativaForm, ConfiguracoesAppForm, EstatisticasAcessoNormasForm)
 from .models import AppConfig, CasaLegislativa
 

sapl/materia/forms.py

@@ -42,7 +42,8 @@ from sapl.utils import (autor_label, autor_modal, timing,
                         models_with_gr_for_model, qs_override_django_filter,
                         SEPARADOR_HASH_PROPOSICAO,
                         validar_arquivo, YES_NO_CHOICES,
-                        GoogleRecapthaMixin, get_client_ip)
+                        GoogleRecapthaMixin)
+from sapl.middleware.ratelimit import get_client_ip
 
 from .models import (AcompanhamentoMateria, Anexada, Autoria,
                      DespachoInicial, DocumentoAcessorio, Numeracao,

sapl/materia/views.py

@@ -54,10 +54,11 @@ from sapl.parlamentares.models import Legislatura
 from sapl.protocoloadm.models import Protocolo
 from sapl.settings import MAX_DOC_UPLOAD_SIZE, MEDIA_ROOT, RATE_LIMITER_RATE
 from sapl.utils import (autor_label, autor_modal, gerar_hash_arquivo, get_base_url,
-                        get_client_ip, get_mime_type_from_file_extension, lista_anexados,
+                        get_mime_type_from_file_extension, lista_anexados,
                         mail_service_configured, montar_row_autor, SEPARADOR_HASH_PROPOSICAO,
                         show_results_filter_set, get_tempfile_dir,
-                        google_recaptcha_configured, MultiFormatOutputMixin, ratelimit_ip)
+                        google_recaptcha_configured, MultiFormatOutputMixin)
+from sapl.middleware.ratelimit import get_client_ip, ratelimit_ip
 
 from .forms import (AcessorioEmLoteFilterSet, AcompanhamentoMateriaForm,
                     AnexadaEmLoteFilterSet, AdicionarVariasAutoriasFilterSet,

sapl/static/robots.txt

@@ -22,6 +22,10 @@ User-agent: SERankingBacklinksBot
 Disallow: /
 Crawl-delay: 10
 
+User-agent: anthropic-python
+Disallow: /
+Crawl-delay: 10
+
 User-agent: *
 Disallow: /relatorios/
 Crawl-delay: 10

sapl/utils.py

@@ -78,6 +78,7 @@ def is_weak_password(password):
     return len(password) < MIN_PASSWORD_LENGTH or not (pwd_has_lowercase and pwd_has_uppercase
                                                        and pwd_has_number and pwd_has_special_char)
 
+
 def groups_remove_user(user, groups_name):
     from django.contrib.auth.models import Group
 
@@ -401,7 +402,6 @@ def xstr(s):
     return '' if s is None else str(s)
 
 
-
 def get_base_url(request):
     # TODO substituir por Site.objects.get_current().domain
     # from django.contrib.sites.models import Site
@@ -1104,6 +1104,7 @@ def cached_call(key, timeout=300):
             return result
 
         return wrap
+
     return cache_decorator
 
 
@@ -1324,7 +1325,7 @@ def get_path_to_name_report_map():
 
 
 class Row:
-    def __init__(self, cols, is_header = False):
+    def __init__(self, cols, is_header=False):
         self.cols = cols
         self.is_header = is_header
 
@@ -1333,7 +1334,7 @@ class Row:
 
 
 class Table:
-    def __init__(self, header = [], rows = []):
+    def __init__(self, header=[], rows=[]):
         self.header = header
         self.rows = rows
 
@@ -1534,7 +1535,7 @@ class MultiFormatOutputMixin:
 
             verbose_name = []
 
-            if hasattr(self, f'hook_header_{fname}'): # suporta extensao de funcionalidade
+            if hasattr(self, f'hook_header_{fname}'):  # suporta extensao de funcionalidade
                 h = getattr(self, f'hook_header_{fname}')()
                 yield h
                 continue

work_queues.md

@@ -0,0 +1,173 @@
+# SAPL — Work Queues & Real-Time: Async PDF + WebSocket Voting
+
+> **Status**: Planned follow-up mini-project.  
+> **Prerequisite**: Redis A (cache + rate-limiter pod, `rate-limiter-2026` branch) must be
+> deployed to production, stable, and OOM pressure confirmed reduced before starting this work.  
+> **Scope**: Django 2.2 / Gunicorn / Celery / Django Channels — same fleet of 1,200+ pods.
+
+---
+
+## Table of Contents
+
+1. [Context & Motivation](#1-context--motivation)
+2. [Redis Topology for Work Queues](#2-redis-topology-for-work-queues)
+3. [Phase 1 — Async PDF via Celery](#3-phase-1--async-pdf-via-celery)
+4. [Phase 2 — Django Channels (WebSocket Voting Panel)](#4-phase-2--django-channels-websocket-voting-panel)
+5. [Open Questions](#5-open-questions)
+
+---
+
+## 1. Context & Motivation
+
+After `rate-limiter-2026` ships:
+
+| Remaining pain point | Current behaviour | Target |
+|---|---|---|
+| PDF generation | Holds a Gunicorn worker thread for the full build duration (10–60 s). Workers are at 400 MB cap — a PDF request burns one slot for up to a minute | Enqueue via Celery; respond 202 immediately; worker is freed |
+| WebSocket voting panel | Not implemented; councillors use a polling page | Persistent connection via Django Channels backed by Redis |
+
+---
+
+## 2. Redis Topology for Work Queues
+
+> **Critical constraint**: Celery broker **must** be a **separate** Redis instance (Redis B)
+> with `noeviction` policy.  
+> Redis A (cache + rate-limiter) uses `allkeys-lru` — tasks enqueued there would be silently
+> evicted under memory pressure, causing jobs to vanish without error.
+
+| Instance | Role | Eviction policy | Persistence |
+|---|---|---|---|
+| **Redis A** (existing) | Page cache (DB0), rate limiter (DB1), Django Channels (DB2) | `allkeys-lru` | none |
+| **Redis B** (new) | Celery broker + result backend | `noeviction` | AOF + RDB snapshot |
+
+```yaml
+# docker/k8s/redis-celery-configmap.yaml
+data:
+  redis.conf: |
+    maxmemory-policy noeviction   # never evict tasks
+    appendonly yes                 # AOF persistence ON
+    save "900 1"                   # RDB snapshot every 15 min if ≥1 change
+    databases 2                    # DB0 = broker queue, DB1 = result backend
+```
+
+---
+
+## 3. Phase 1 — Async PDF via Celery
+
+### 3.1 Current (synchronous) flow
+
+Holds worker memory for the entire PDF build:
+
+```mermaid
+sequenceDiagram
+    participant B as Browser
+    participant G as Gunicorn worker
+    participant ORM as PostgreSQL
+    participant RL as ReportLab
+
+    B->>G: GET /pdf/materia/12345
+    G->>ORM: N+1 queries (get_etiqueta_protocolos)
+    ORM-->>G: data
+    G->>RL: build entire PDF in RAM
+    RL-->>G: PDF bytes (held in worker memory)
+    G-->>B: stream response
+    note over G: worker blocked for full duration
+```
+
+### 3.2 Target (async) flow
+
+Worker freed immediately after enqueueing:
+
+```mermaid
+sequenceDiagram
+    participant B as Browser
+    participant G as Gunicorn worker
+    participant Q as Redis B (Celery queue)
+    participant W as Celery worker
+    participant D as Disk / nginx
+
+    B->>G: POST /pdf/materia/12345
+    G->>Q: enqueue task
+    G-->>B: 202 Accepted + task_id
+    W->>W: build PDF (out of band)
+    W->>D: write PDF to /media/pdf/task_id.pdf
+    B->>G: GET /pdf/status/task_id
+    G-->>B: 302 → nginx /media/pdf/task_id.pdf
+```
+
+### 3.3 Celery settings
+
+```python
+# sapl/settings.py additions
+CELERY_BROKER_URL     = config('CELERY_BROKER_URL',     default='')
+CELERY_RESULT_BACKEND = config('CELERY_RESULT_BACKEND', default='')
+
+# Soft limit: warn at 350 MB; hard limit: kill+restart at 450 MB.
+# Keeps Celery workers inside the same memory envelope as Gunicorn workers.
+CELERY_WORKER_MAX_MEMORY_PER_CHILD = 400 * 1024  # KB
+CELERY_TASK_SOFT_TIME_LIMIT = 120   # seconds — warn
+CELERY_TASK_TIME_LIMIT      = 180   # seconds — SIGKILL
+```
+
+### 3.4 k8s manifests
+
+New files to be created under `docker/k8s/`:
+
+- `redis-celery-configmap.yaml` — Redis B config (noeviction, AOF)
+- `redis-celery-deployment.yaml` — single-replica Redis B pod
+- `redis-celery-service.yaml` — ClusterIP service
+- `celery-deployment.yaml` — Celery worker deployment (same image as SAPL)
+
+### 3.5 Environment variables (per-namespace Secret)
+
+| Variable | Example value | Notes |
+|---|---|---|
+| `CELERY_BROKER_URL` | `redis://sapl-redis-celery.redis.svc:6379/0` | Redis B, DB0 |
+| `CELERY_RESULT_BACKEND` | `redis://sapl-redis-celery.redis.svc:6379/1` | Redis B, DB1 |
+
+---
+
+## 4. Phase 2 — Django Channels (WebSocket Voting Panel)
+
+Uses **Redis A DB2** (reserved in the existing key-layout table — no new infra needed beyond
+what ships in `rate-limiter-2026`).
+
+### 4.1 Channel layer settings
+
+```python
+# sapl/settings.py additions
+CHANNEL_LAYERS = {
+    "default": {
+        "BACKEND": "channels_redis.core.RedisChannelLayer",
+        "CONFIG": {
+            "hosts": [("sapl-redis.redis.svc.cluster.local", 6379)],
+            "db": 2,        # DB2 reserved for channels (see rate-limiter-v2.md §0.2)
+            "capacity": 1500,
+            "expiry": 10,
+        },
+    }
+}
+```
+
+### 4.2 Prerequisites before starting
+
+- [ ] Redis A stable in production (rate limiter + cache confirmed working)
+- [ ] OOM kill rate reduced to near-zero
+- [ ] Bot siege resolved (Phase 0–2 metrics reviewed)
+- [ ] Decision on ASGI server (Daphne vs Uvicorn + channels) — Gunicorn alone cannot serve WebSockets
+
+---
+
+## 5. Open Questions
+
+| # | Question | Blocks |
+|---|---|---|
+| 1 | Which PDF endpoints are highest priority for async migration? (`/relatorios/`, `/materia/pdf/`, other)? | Phase 1 scope |
+| 2 | Should the Celery worker run in the same pod as Gunicorn (sidecar) or a dedicated deployment? | Phase 1 k8s design |
+| 3 | Result backend TTL — how long should generated PDFs be retained before cleanup? | Phase 1 storage design |
+| 4 | ASGI server selection for Channels (Daphne vs uvicorn + channels) | Phase 2 |
+| 5 | WebSocket voting panel: is per-session or per-pod state acceptable? | Phase 2 architecture |
+
+---
+
+*Planned work — begins after `rate-limiter-2026` is stable in production.*