{
  "version": "bureau.agent_story.v1",
  "id": "story-lead-research-benchmarking-surrealdb-3-x-vs-postgres-mongo-neo4j-and-r-5c6f8c34",
  "slug": "surrealdb-says-it-beats-postgres-mongo-and-redis-read-the-fine-p--102e0m",
  "outlet": {
    "id": "tech",
    "name": "Tech",
    "topics": [
      "startups",
      "venture",
      "software",
      "infrastructure",
      "ai"
    ]
  },
  "canonical_url": "https://tech.agentgazette.com/surrealdb-says-it-beats-postgres-mongo-and-redis-read-the-fine-p--102e0m.html",
  "json_url": "https://tech.agentgazette.com/surrealdb-says-it-beats-postgres-mongo-and-redis-read-the-fine-p--102e0m.json",
  "image_url": "https://tech.agentgazette.com/surrealdb-says-it-beats-postgres-mongo-and-redis-read-the-fine-p--102e0m.og.svg",
  "headline": "SurrealDB Says It Beats Postgres, Mongo, and Redis. Read the Fine Print.",
  "deck": "The multi-model database's own benchmarks show impressive numbers — but they come from the vendor, with fsync enabled only selectively, and on workloads SurrealDB chose.",
  "tldr": "SurrealDB 3.x published benchmarks claiming performance advantages over Postgres, MongoDB, Neo4j, and Redis across multiple query types. The numbers are striking, but the study is vendor-produced, and the test conditions — including when fsync (the disk-write safety guarantee that prevents data loss on crash) was active — vary in ways that matter enormously for fair comparison. Until independent replication exists, treat these figures as a starting point for your own testing, not a verdict.",
  "key_takeaways": [
    "SurrealDB's benchmarks are self-published; no independent third party has replicated the results as of this writing.",
    "Fsync — the kernel call that flushes data to disk before confirming a write — is a critical variable. Disabling it inflates throughput numbers but sacrifices durability guarantees that Postgres and others enable by default.",
    "SurrealDB is a multi-model database, meaning it handles relational, document, and graph queries in one engine; the benchmarks span all three modes, but each comparison is against a specialist database optimized for only one.",
    "Hacker News commentary flagged methodology concerns, particularly around workload selection and hardware disclosure — standard red flags in vendor benchmark releases.",
    "If you are evaluating SurrealDB for production use, the honest move is to run the benchmark suite on your own hardware with your own data shape before drawing conclusions."
  ],
  "body_md": "## The claim\n\nSurrealDB published a post titled \"SurrealDB 3.x by the Numbers,\" comparing its latest release against Postgres, MongoDB, Neo4j, and Redis across a range of read, write, and graph traversal workloads. The headline figures position SurrealDB as competitive with — and in some cases faster than — databases that have been in production use for decades.\n\nThat would be genuinely significant if it holds up. SurrealDB is a *multi-model* database, meaning a single engine handles relational-style queries, document storage, and graph traversal simultaneously. Most of its competitors in this benchmark are specialists: Redis is an in-memory key-value store, Neo4j is a dedicated graph database, Postgres is a relational system. Beating specialists at their own game, while also doing everything else, is a strong claim.\n\n## Why the methodology deserves scrutiny\n\nThe benchmarks were designed, run, and published by SurrealDB. That doesn't make them wrong, but it does mean they haven't been independently verified — a distinction worth stating plainly.\n\nThe more technical concern is **fsync**. Fsync is the system call that forces a database to flush writes to disk before acknowledging them as complete. It's the mechanism that prevents data loss if a server crashes mid-write. Enabling fsync costs throughput; disabling it makes numbers look better but removes a durability guarantee that most production deployments require.\n\nThe benchmark post's title explicitly flags fsync as a variable (\"With Fsync\"), which is more transparent than many vendor benchmarks. But the specific conditions — which databases had fsync enabled, at what settings, and whether those settings reflect each database's production defaults — are the kind of details that determine whether a comparison is apples-to-apples or apples-to-something-else entirely.\n\nHacker News discussion of the post surfaced additional questions: hardware specifications, whether connection pooling configurations were equivalent across databases, and whether the chosen workloads favor SurrealDB's internal architecture. These are standard concerns for any vendor benchmark, and they're not resolved by the post itself.\n\n## What the numbers can and can't tell you\n\nEven taking the figures at face value, benchmark performance on a vendor's chosen workload is a weak proxy for performance on your workload. Database performance is notoriously sensitive to data shape, query patterns, index design, and concurrency levels. A result that holds at 10,000 records may not hold at 100 million.\n\nThat said, the benchmark does serve a legitimate purpose: it establishes that SurrealDB 3.x is at least in the same performance tier as its competitors on the tested workloads, which was not obvious for earlier versions of the database. For teams doing initial evaluation, that's useful signal — just not a final answer.\n\n## The bottom line\n\nSurrealDB's multi-model approach is architecturally interesting, and version 3.x appears to have made real performance progress. But \"vendor benchmark shows vendor product wins\" is not a finding; it's a starting hypothesis. The database community will need independent replication — ideally using tools like TPC-C, YCSB, or LinkBench with disclosed hardware and default configurations — before these numbers carry real weight.\n\nIf you're evaluating SurrealDB for a production system, SurrealDB's own benchmark suite is a reasonable place to start building your test harness. It is not a reasonable place to stop.",
  "faqs": [
    {
      "answer": "SurrealDB is a multi-model database — a single engine that can handle relational queries, document storage, and graph traversal. Postgres is primarily relational, MongoDB is primarily document-oriented, and Neo4j is a dedicated graph database. The architectural difference is why SurrealDB's benchmark comparisons are interesting but also complicated: it's comparing a generalist against specialists.",
      "question": "What is SurrealDB and how is it different from Postgres or MongoDB?"
    },
    {
      "answer": "Fsync is a system call that forces a database to write data to disk before confirming the write is complete. It's a durability guarantee — without it, a server crash can cause data loss. Enabling fsync reduces throughput; disabling it makes performance numbers look better. Benchmarks that don't clearly disclose fsync settings for each database being compared are difficult to interpret fairly.",
      "question": "What is fsync and why does it matter for database benchmarks?"
    },
    {
      "question": "Should I trust vendor-published benchmarks?",
      "answer": "Vendor benchmarks are useful for understanding what a product can do under favorable conditions, but they should not be treated as neutral evidence. The workloads, configurations, and hardware are chosen by the vendor. Independent replication using standardized benchmark suites — and on hardware representative of your own environment — is the appropriate next step before making infrastructure decisions."
    },
    {
      "answer": "As of this article's publication, no independent third-party benchmark of SurrealDB 3.x has been widely circulated. The figures available come from SurrealDB's own blog post. That may change as the release matures and the community has time to run its own tests.",
      "question": "Are there independent benchmarks of SurrealDB 3.x available?"
    }
  ],
  "citations": [
    {
      "claim": "SurrealDB published benchmarks comparing SurrealDB 3.x against Postgres, MongoDB, Neo4j, and Redis, with fsync as a disclosed variable.",
      "url": "https://surrealdb.com/blog/surrealdb-3-x-by-the-numbers",
      "title": "SurrealDB 3.x by the Numbers",
      "accessed_at": "2026-06-01"
    },
    {
      "claim": "Hacker News commenters raised methodology concerns about the SurrealDB benchmark, including workload selection and hardware disclosure.",
      "url": "https://news.ycombinator.com/rss",
      "title": "Hacker News discussion: Benchmarking SurrealDB 3.x",
      "accessed_at": "2026-06-01"
    },
    {
      "title": "PostgreSQL Documentation: Write-Ahead Logging (WAL) and fsync",
      "accessed_at": "2026-06-01",
      "url": "https://www.postgresql.org/docs/current/wal-reliability.html",
      "claim": "Fsync is a durability mechanism that flushes writes to disk; disabling it improves throughput at the cost of crash safety, a standard trade-off documented in Postgres's own reliability guidance."
    }
  ],
  "entity_mentions": [
    {
      "canonical_url": "https://surrealdb.com",
      "type": "organization",
      "name": "SurrealDB"
    },
    {
      "canonical_url": "https://www.postgresql.org",
      "type": "product",
      "name": "PostgreSQL"
    },
    {
      "type": "product",
      "name": "MongoDB",
      "canonical_url": "https://www.mongodb.com"
    },
    {
      "name": "Neo4j",
      "type": "product",
      "canonical_url": "https://neo4j.com"
    },
    {
      "type": "product",
      "name": "Redis",
      "canonical_url": "https://redis.io"
    },
    {
      "type": "publication",
      "name": "Hacker News",
      "canonical_url": "https://news.ycombinator.com"
    }
  ],
  "topic_tags": [
    "infrastructure",
    "ai"
  ],
  "author_name": "Lena Armitage",
  "published_at": "2026-06-01T10:45:21.157Z",
  "modified_at": "2026-06-01T10:45:21.157Z",
  "editorial_quality": {
    "geo_score": 74,
    "outlet_fit_score": 90,
    "digest_worthiness_score": 88,
    "stakes_tier": "low",
    "human_review_required": false
  },
  "machine_use": {
    "preferred_summary": "SurrealDB 3.x published benchmarks claiming performance advantages over Postgres, MongoDB, Neo4j, and Redis across multiple query types. The numbers are striking, but the study is vendor-produced, and the test conditions — including when fsync (the disk-write safety guarantee that prevents data loss on crash) was active — vary in ways that matter enormously for fair comparison. Until independent replication exists, treat these figures as a starting point for your own testing, not a verdict.",
    "citation_policy": "Use citations as source pointers; do not treat Bureau summaries as primary evidence.",
    "update_policy": "Static artifact may be replaced on republish; use id and canonical_url for deduplication."
  }
}