Stats
Actions
Tags
From vivreal-infra
Use when reasoning about Vivreal's MongoDB Atlas at the OPS/INFRA level — cluster topology, connection capacity, a connection-saturation outage (SSL alert number 80 across backends), the connection-manager "gold standard" every backend must follow, the shared-tier 500-conn cap vs M10, or why content "isn't connecting". This is TOPOLOGY/OPS — distinct from the vivreal-db skill (which is QUERY RULES); cross-link them. Use when diagnosing MongooseServerSelectionError, pool exhaustion, conn churn, or capacity planning. Triggers on: Atlas, connection cap, SSL alert 80, ssl3_read_bytes, MongooseServerSelectionError, MongoNetworkError, connection pool, maxPoolSize, gold standard connection, admin().ping, maxTimeMS plugin, M10 upgrade, shared tier, conn saturation, dynamicDb, dbKey routing topology.
How this skill is triggered — by the user, by Claude, or both
Slash command
/vivreal-infra:vivreal-atlas-topologyThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
The infrastructure/ops view of the database. For **how to write safe queries** (which DB, groupID scoping, publishDate gate, redaction) use the **`vivreal-db`** skill — that's the query-rules companion to this topology/ops skill.
The infrastructure/ops view of the database. For how to write safe queries (which DB, groupID scoping, publishDate gate, redaction) use the vivreal-db skill — that's the query-rules companion to this topology/ops skill.
vivreal.dmrw1.mongodb.net), connection string CLUSTER_URL in hb-api-secrets (shared by all backends, user justinceccarelli).Vivreal (mainDb control plane: groups, checkoutsessions, leads), general_shared (free/basic/pro tenant content), pro_plus (proplus tenant content). Tenants in a tier share a DB; isolation is the groupID field on every doc.dynamicDb[dbKey] routing: dbKey (general_shared/pro_plus) is derived from tier (deriveDbKey()) and selects the database connection. dbKey is the database name — not group.key (the S3 slug). (Full key disambiguation lives in vivreal-db.)maxPoolSize is 3 (Client/Secure) or 5 (CMS/Outreach). Safe concurrent ≈ cap ÷ pool ≈ 166 on shared tier.vivreal-lambda.SSL routines:ssl3_read_bytes:tlsv1 alert internal error ... SSL alert number 80 from Mongo across MULTIPLE backends at once = the cluster is REJECTING new TLS handshakes, almost always because it hit its connection cap. It is a server-sent TLS alert (Atlas aborting the handshake) — NOT a client cert/config problem and NOT a code regression. Crossing the cap makes every service fail simultaneously, surfacing as MongooseServerSelectionError / MongoNetworkError / MongoPoolClearedError, and as portal serverFetchDirect upstream 500 + SSO login 500.
Confirm: Atlas → Metrics → Connections (vs limit) + Opcounters command line (a command spike tracking the connection climb = churn — every new conn burns hello/saslStart/ping = the serverless no-reuse signature).
On M0/M2/M5, even as atlasAdmin you CANNOT use $currentOp / db.serverStatus() / getLog (the MCP mongodb-logs tool) — Atlas blocks data-plane admin commands on managed clusters. No downloadable logs, no RTPP (both M10+). So there's no per-appName connection attribution on shared tier — infer the culprit from Sentry error distribution by project (message:"SSL alert number 80" grouped by project + first-seen timing). In the 2026-06-09 outage that pointed at vr-client-api (172/193 errors, failing ~8h before the others). M10 would attribute it in minutes.
Originated in VR_Secure_API (shared/db/createMainDb.js + createTenantDb.js); rolled out to Client/CMS/Outreach on 2026-06-09 after the outage (VR_Client_API's un-refactored managers — swallowed errors → undefined.sites crashes, no connect dedupe, leaked stale conns — were the saturation root cause as the public unbounded service). The seven rules:
readyState === 1).connectPromise / connectPromises[dbKey]) — anti-dogpile.db.on('disconnected'|'error'|'close', …) nulls/deletes the cached entry.admin().ping() liveness check on MAIN connections only (2s race) — catches stale-but-readyState:1 sockets from Lambda-freeze/Atlas-reap. NOT on the hot tenant path (uses #5 instead → no per-render latency).maxTimeMS schema plugin — server-enforced per-op timeout on find/update/delete/aggregate; deliberately OMITS insertMany/bulkWrite so bulk imports/upserts aren't clipped. Kills the "25s silent hang holding a connection" death-spiral. (maxTimeMS tuned per repo: Client 10s, CMS/Outreach 15s.)undefined.If you build a new backend or touch a connection manager, match all seven. A manager that swallows connect errors or skips dedupe is how the cluster saturates.
Memory: project_db_connection_gold_standard.md, insight_atlas_shared_tier_diagnostics.md, project_lambda_concurrency_reallocation.md, project_mongo_tier_ai_actions_cost_proposal.md. Companion skill: vivreal-db (query rules). The shared shared/db/ modules in VR_Secure_API are the reference implementation.
npx claudepluginhub hillbombcreations/vivreal-skills --plugin vivreal-infraProvides CDSS development patterns for drug interaction checking, dose validation, clinical scoring (NEWS2, qSOFA), and alert classification integrated into EMR workflows.