我的书签
添加书签
移除书签Muse Spark 1.2·续
Head-of-Line Blocking & Hot Partitions.
Because ordering is strict per-partition, a single poison pill event or a high-volume customer will block all subsequent events for every other customer mapped to that same partition. The consumer cannot skip the failed event without violating order.
Mitigation required: 1) Idempotent consumers for at-least-once restarts, 2) Dead-letter topic with non-blocking retry to park poison pills after N attempts, and 3) Watch for partition skew - split hot customers with virtual partitions or salting if one customer dominates throughput, otherwise lag will grow unbounded.
```json
{
"id": "resp_6ac7b061a100161e1ff649a1",
"object": "response",
"created_at": 1791471713,
"model": "muse-spark-1.2",
"output": [
{
"type": "reasoning",
"id": "rs_6ac7b061a100161e1ff649a1:rs_60a314bf-06b8-48e8-a780-22fd4ffef97f",
"summary": [],
"status": "completed"
},
{
"type": "message",
"id": "msg_18082550-8bb3-4b52-a537-4673ef8152ab",
"role": "assistant",
"content": [
{
"type": "output_text",
"text": "For 50k events/sec with strict ordering + replay, **Partitioned Log** is the correct pattern.\n\n### Decision Table: Partitioned Log vs Task Queue\n\n| Criteria | Partitioned Log `Kafka / Redpanda / Kinesis` | Task Queue `SQS / RabbitMQ / Celery / Sidekiq` |\n| :--- | :--- | :--- |\n| **Ordering** | **Strong per-key ordering.** Partition key = `customer_id` guarantees all events for one customer go to one ordered partition. Total order not guaranteed, but per-customer order is. | **Weak / No ordering.** Most queues provide best-effort FIFO per-queue, but scaling with multiple consumers breaks ordering. Requires one queue per customer to guarantee order - not feasible. |\n| **Throughput** | **High & Horizontally Scalable.** 50k/sec is easily handled by adding partitions and brokers. Sequential disk I/O. Consumers scale with partition count. | **Limited by broker/acknowledgement overhead.** High throughput requires many queues/workers. Contention on central broker, prefetch limits and per-message ACKs reduce throughput at 50k/sec. |\n| **Retries** | **At-least-once, consumer-managed.** Offset not committed until processed. On restart, consumer resumes from last committed offset. Requires idempotent consumer + Dead Letter Topic for poison pills to avoid blocking partition. | **Built-in ACK/NACK & retries.** Failed job NACKs and is retried/requeued automatically. Easy per-message retry with visibility timeout / backoff. Risk of duplicate execution if worker crashes after processing but before ACK. |\n| **Replay** | **Native, for retention window.** Log is durable and immutable. Any consumer can rewind offset and replay last 7 days by configuration `retention.ms=7 days`. Ideal for new consumers, debugging, and reprocessing. | **No replay.** Once ACKed, message is deleted. Requires separate archival DB to rebuild history. Cannot replay 7 days without custom infrastructure. |\n| **Operations** | **Higher operational cost.** Must manage brokers, partitions, replication, consumer rebalancing, schema registry, and monitor consumer lag. Partition count is hard to change and hot customers create skew. | **Lower operational cost.** Fully managed services `SQS` are simple. No partitions/rebalancing to manage. Monitoring is queue depth. |\n\n### Recommendation: Partitioned Log\n\nA Task Queue cannot meet two core requirements: per-customer ordering at scale and 7-day replay. It is designed for *distribution of work*, not *ordered event history*.\n\nA Partitioned Log is designed for this exact workload. Partition by `customer_id` to preserve ordering, scale consumers via a consumer group `1 consumer per partition`, set retention to 7 days, and use committed offsets to tolerate restarts.\n\n### Main Failure Mode to Design For\n\n**Head-of-Line Blocking & Hot Partitions.**\n\nBecause ordering is strict per-partition, a single poison pill event or a high-volume customer will block all subsequent events for every other customer mapped to that same partition. The consumer cannot skip the failed event without violating order.\n\n**Mitigation required:** 1) Idempotent consumers for at-least-once restarts, 2) Dead-letter topic with non-blocking retry to park poison pills after N attempts, and 3) Watch for partition skew - split hot customers with virtual partitions or salting if one customer dominates throughput, otherwise lag will grow unbounded.",
"annotations": [],
"logprobs": []
}
],
"status": "completed"
}
],
"status": "completed",
"usage": {
"input_tokens": 80,
"output_tokens": 1386,
"total_tokens": 1466,
"input_tokens_details": {
"cached_tokens": 0
},
"output_tokens_details": {
"reasoning_tokens": 667
}
},
"top_logprobs": 0,
"temperature": 1,
"top_p": 1,
"service_tier": "auto",
"reasoning": {
"effort": "high"
},
"background": false,
"tools": [],
"tool_choice": "auto",
"truncation": "disabled",
"completed_at": 1791471713,
"error": null,
"incomplete_details": null,
"instructions": null,
"parallel_tool_calls": true,
"store": true,
"max_output_tokens": 4096
}
**Web Search Grounding** — Use Meta's Responses API web search grounding to answer a current-events question with citations.
const response = await env.AI.run(
'meta/muse-spark-1.2',
{
input: 'What is the latest major technology news today? Cite the sources.',
tools: [{ type: 'web_search' }],
},
)
console.log(response)
curl https://api.cloudflare.com/client/v4/accounts/$CLOUDFLARE_ACCOUNT_ID/ai/run \
--header "Authorization: Bearer $CLOUDFLARE_API_TOKEN" \
--header "Content-Type: application/json" \
--data '{
"model": "meta/muse-spark-1.2",
"input": {
"input": "What is the latest major technology news today? Cite the sources.",
"tools": [{"type": "web_search"}]
}
}'
I'll gather today's biggest tech headlines for you — searching the latest coverage now.
{
"id": "resp_6ac7b061a0901a5463764dc7",
"object": "response",
"created_at": 1791471713,
"model": "muse-spark-1.2",
"output": [
{
"type": "reasoning",
"id": "rs_6ac7b061a0901a5463764dc7:rs_2c2099d2-8863-470a-8810-2ddf2fff9d3e",
"summary": [],
"status": "completed"
},
{
"type": "message",
"id": "rs_cb639230-45c1-4b26-a436-faaf8ab36206",
"role": "assistant",
"content": [
{
"type": "output_text",
"text": "I'll gather today's biggest tech headlines for you — searching the latest coverage now.",
"annotations": [],
"logprobs": []
}
],
"phase": "commentary",
"status": "completed"
},
{
"type": "web_search_call",
"id": "ws_call_1490572f-13ac-4577-ba12-f6f4907064ee",
"status": "completed",
"action": {
"type": "search",
"query": "major technology news today October 8 2026"
}
},
{
"type": "web_search_call",
"id": "ws_call_4744e507-abd8-4bb6-b4a3-5fb5c178ceb4",
"status": "completed",
"action": {
"type": "search",
"query": "tech news headlines October 8 2026"
}
},
{
"type": "web_search_call",
"id": "ws_call_40146bf5-4464-4d1e-9427-31493db38718",
"status": "completed",
"action": {
"type": "search",
"query": "breaking technology news October 2026"
}
},
{
"type": "web_search_call",
"id": "ws_call_f37bd93c-4f72-4728-9b31-afd646c47129",
"status": "completed",
"action": {
"type": "search",
"query": "\"October 8, 2026\" technology"
}
},
{
"type": "web_search_call",
"id": "ws_call_ecaabafc-bbeb-4527-963d-865fb5de9357",
"status": "completed",
"action": {
"type": "open_page",
"url": "https://www.archynetys.com/trend/2026-10-08/samsung-forecasts-record-third-quarter-profit-of-80-billion-on-the-back-of-ai"
}
},
{
"type": "web_search_call",
"id": "ws_call_7fdbfebc-908c-4add-96eb-742232c1accf",
"status": "completed",
"action": {
"type": "open_page",
"url": "https://www.archynetys.com/trend/2026-10-07/microsoft-surface-and-windows-event-2026-all-the-announcements-and-analysis"
}
},
{
"type": "web_search_call",
"id": "ws_call_38245068-43bb-4a86-baa8-0731d239533e",
"status": "completed",
"action": {
"type": "open_page",
"url": "https://techstartups.com/2026/10/08/startup-funding-news-today-october-8-2026-manus-universal-quantum-mecka-verso-more/"
}
},
{
"type": "web_search_call",
"id": "ws_call_bd19abc4-53ab-49a7-bb7f-b42a6f7ebaeb",
"status": "completed",
"action": {
"type": "search",
"query": "technology news today AI chip"
}
},
{
"type": "web_search_call",
"id": "ws_call_db54ecc1-ee6a-4f22-83f8-5ba42c60606b",
"status": "completed",
"action": {
"type": "search",
"query": "latest technology news Reuters AP October 8 2026"
}
},
{
"type": "web_search_call",
"id": "ws_call_581890ae-b3e4-4e91-81d1-a41b60a5d869",
"status": "completed",
"action": {
"type": "search",
"query": "The Verge tech news today October 8 2026"
}
},
{
"type": "web_search_call",
"id": "ws_call_1d4fa484-1ac1-4e9a-a7ed-d7559ff906de",
"status": "completed",
"action": {
"type": "search",
"query": "Samsung profit record AI demand October 8 2026"
}
},
{
"type": "web_search_call",
"id": "ws_call_c3f9d6b5-5943-423a-9b36-43ddf5ed8b21",
"status": "completed",
"action": {
"type": "open_page",
"url": "https://www.lapost.com/content/samsung-q3-profit-jumps-783-as-ai-memory-boom-lifts-chip-earnings"
}
},
{
"type": "web_search_call",
"id": "ws_call_f630d925-7941-4b83-bec4-baaf78ab778e",
"status": "completed",
"action": {
"type": "open_page",
"url": "https://harpianews.com/ai-chip-growth-pushes-samsung-earnings-to-file-80bn/"
}
},
{
"type": "web_search_call",
"id": "ws_call_42a04b68-1ce3-4369-a913-124d609dba29",
"status": "completed",
"action": {
"type": "open_page",
"url": "http://dailyiq.me/stock/ORCL"
}
},
{
"type": "web_search_call",
"id": "ws_call_9737abf0-ee83-4855-90ad-7b7bf7fd5cd6",
"status": "completed",
"action": {
"type": "open_page",
"url": "https://aiagentstore.ai/ai-agent-news/today"
}
},
{
"type": "web_search_call",
"id": "ws_call_99bdfd45-639e-467e-bd33-279755796cf7",
"status": "completed",
"action": {
"type": "search",
"query": "Wi-Fi 8 routers ASUS TP-Link October 2026"
}
},
{
"type": "reasoning",
"id": "rs_6ac7b061a0901a5463764dc7:rs_4a9e895b-ecc8-41ec-9083-205c72d5dcc4",
"summary": [],
"status": "completed"
},
{
"type": "message",
"id": "msg_bd35a465-b38f-472b-bf73-f82c3256caaf",
"role": "assistant",
"content": [
{
"type": "output_text",
"text": "Latest Major Technology News - Today, October 8, 2026:\n\n**1. Samsung forecasts record Q3 profit on AI memory boom**\n* Samsung Electronics on Thursday projected its quarterly profit would top 100 trillion won, a world first for a technology company, estimating a nearly nine-fold jump in third-quarter earnings as booming demand for AI chips drove strong memory sales.\n* The company projects a $80 billion Q3 profit as AI‑driven chip demand spikes, setting a new industry benchmark, published October 8, 2026 12:38 UTC.\n* Details: Samsung Electronics says it expects a nine-fold surge in its quarterly earnings contrast with a year earlier, pushed by surging demand for memory chips utilized in artificial intelligence (AI) information centres, with operating profit for the three months to the end of September to soar to 107.4tn won ($80bn), slightly ahead of an LSEG SmartEstimate of 106.1 trillion won. Related coverage notes it as Samsung flags $80 billion profit on AI boom, highest quarterly for any tech company and that Samsung’s Q3 2026 profit jumps 782% to $80 billion on AI chip demand. The forecast marks Samsung's fourth straight quarter of record operating profit, driven by AI‑optimized memory chips, described as an AI chip boom.\n\n**2. Microsoft Surface and Windows Event 2026**\n* Microsoft Surface and Windows Event 2026: All the Announcements and Analysis Live From San Francisco — published October 8, 2026 06:46 UTC.\n* Primary Driver: Microsoft unveils MacBook Pro competitor at Windows and Surface event.\n* Microsoft has announced a MacBook Pro competitor, a new line of laptops designed to challenge Apple's dominance in the premium laptop market, with coverage from The Verge explaining that Microsoft built its new laptop with a focus on performance and portability, featuring a custom-designed processor and a sleek, aluminum design.\n\n**3. Oracle - AI chip financing and other tech developments**\n* On Oct 8, Oracle was reported to be negotiating with Apollo and Goldman Sachs to finance a large purchase of AI chips through a lease‑structured vehicle, a move designed to avoid adding to its bond debt amid concerns over rising leverage.\n* Oracle entered negotiations with Apollo and Goldman Sachs to finance a substantial purchase of AI chips through a lease‑structured vehicle, reflecting a broader trend of pursuing private debt arrangements worth tens of billions to fund AI hardware acquisitions.\n\n**4. Venture Funding - AI Agents, Quantum and Robotics dominate Oct 8**\n* Startup funding over the past 24 hours has been dominated by a remarkable comeback in the AI agent market, alongside
... response truncated ...
Parameters
▶input
one ofrequired
instructions
string
temperature
numberminimum: 0maximum: 2
max_output_tokens
numberexclusiveMinimum: 0
top_p
numberminimum: 0maximum: 1
stream
boolean
▶tools[]
array
tool_choice
▶text{}
object
▶reasoning{}
object
id
string
object
stringconst: response
created_at
number
model
string
▶output[]
array
output_text
string
status
stringenum: in_progress, completed, failed, incomplete
▶usage{}
object
API Schemas (Raw)
Was this helpful?
YesNo
On this page
{"@context":"https://schema.org","@type":"TechArticle","@id":"https://developers.cloudflare.com/ai/models/meta/muse-spark-1.2/#page","headline":"Muse Spark 1.2","description":"Muse Spark 1.2 is Meta's agentic and coding model for multi-step tool loops, software engineering assistants, and long-context reasoning.","url":"https://developers.cloudflare.com/ai/models/meta/muse-spark-1.2/","inLanguage":"en","image":"https://developers.cloudflare.com/og-docs.png","publisher":{"@type":"Organization","name":"Cloudflare","description":"One platform for your apps, agents, and workforce. Build, secure, and scale without managing infrastructure","url":"https://www.cloudflare.com/","sameAs":["https://github.com/cloudflare","https://www.linkedin.com/company/cloudflare","https://x.com/cloudflare"],"logo":{"@type":"ImageObject","url":"https://developers.cloudflare.com/logo.svg"},"address":{"@type":"PostalAddress","streetAddress":"101 Townsend St","addressLocality":"San Francisco","addressRegion":"CA","postalCode":"94107","addressCountry":"US"},"contactPoint":[{"@type":"ContactPoint","contactType":"Customer Support","url":"https://support.cloudflare.com/","availableLanguage":["English"]},{"@type":"ContactPoint","contactType":"Sales","url":"https://www.cloudflare.com/contact/","availableLanguage":["English"]}]},"isPartOf":{"@type":"WebSite","@id":"https://developers.cloudflare.com/#website","name":"Cloudflare Docs","url":"https://developers.cloudflare.com/"}}
description: Muse Spark 1.3 is Meta's latest agentic and coding model, tuned for multi-step tool loops, browser tasks, software engineering, and long-horizon reasoning. title: Muse Spark 1.3 image: https://developers.cloudflare.com/og-docs.png
Documentation Index
Fetch the complete documentation index at: https://developers.cloudflare.com/llms.txt
Use this file to discover all available pages before exploring further.
