> ## Documentation Index > Fetch the complete documentation index at: https://docs.vortexiq.ai/llms.txt > Use this file to discover all available pages before exploring further. # Apache Cassandra on Vortex IQ > Monitor Apache Cassandra health, cost and reliability signals, and catch incidents and runaway spend early. Monitor Apache Cassandra health, cost and reliability signals, and catch incidents and runaway spend early. [Connect or manage this source](https://app.vortexiq.ai/workbench/settings/sources) · [How connecting works](/integrations/connector-catalogue) · [Create a workflow](https://app.vortexiq.ai/workbench/flows/create?connector=cassandra) performance signals automated checks prepared fixes proven workflows API operations ### What you can achieve Capabilities are grouped around merchant outcomes, not API terminology. Monitor orders, fulfilment, delivery and settlement. Find failures, leaks and risks before they cost sales. Find storefront, speed, accessibility and journey problems. Improve discovery, conversion, campaigns and repeat purchase. Keep tracking, access and change under governed control. ### From connection to verified outcome The controlled sequence every capability follows. Nothing changes a connected system without the approval step. Authorise the source. Scopes are shown before access is granted. Watch the signals against your own baselines, not universal defaults. Run checks and gather evidence specific to your store. Explain what happened, why it matters and the proposed action. You review scope, risk and reversibility before anything changes. Apply through governed connector operations. Confirm the intended result and keep the receipt. No changes are made without the configured approval policy. Read-only operations do not modify the connected system; schedules, access scopes, API usage and data handling remain governed by Vortex IQ controls. ### Monitor performance 29 performance signals. Open an outcome to see its signals and how each one alerts. Read-only operations do not modify the connected system. | Signal | Alert behaviour | What it tracks | | ---------------------------------------------- | ------------------ | ------------------------------------------------------------------------------------------------------------------------------------------------------- | | **Cassandra Health Score** | Merchant rule | Composite of node-up ratio, latency percentiles, pending compactions and dropped mutations. | | **Disk Usage %** | Alert band 70 / 90 | Load / data-directory usage from nodetool status Load column vs disk capacity. | | **Hinted Handoffs Pending** | Merchant rule | Cassandra-distinctive - hints stored for unreachable replicas. Growing = a replica is down or slow; data not yet converged. | | **JVM Heap Used %** | Merchant rule | Cassandra runs on the JVM - heap >75% drives GC pauses; sustained high heap precedes node instability. | | **Key Cache Hit Rate %** | Alert band 95 / 80 | KeyCache hit\_ratio from the system\_views.caches CQL virtual table (4.0+) / org.apache.cassandra.metrics:type=Cache,scope=KeyCache,name=HitRate MBean. | | **Native-Transport Connections** | Watch only | Connected client count from system\_views.clients. | | **Native-Transport Pool Saturation %** | Alert band 70 / 90 | Native-Transport-Requests active / max threads from system\_views.thread\_pools. | | **Node Down (DN) or Gossip Flapping** | Merchant rule | Cassandra-distinctive - a DN node reduces replica availability; sustained = consistency-level failures imminent. Page on-call. | | **Nodes Up / Normal (UN)** | Merchant rule | Count of endpoints in UN state from nodetool status / system.peers. Cassandra-defining - DN node = reduced replica availability. | | **Operations per Second (read + write, live)** | Watch only | ClientRequest Read + Write count delta from system\_views / org.apache.cassandra.metrics MBeans. | | **Pending Compactions** | Merchant rule | Cassandra-distinctive - from nodetool compactionstats / PendingTasks MBean. Backlog = SSTables stacking up, read amplification rising. | | **SSTables per Read (avg)** | Merchant rule | SSTablesPerReadHistogram - high count = compaction falling behind, every read touches more files. | | Signal | Alert behaviour | What it tracks | | ------------------------------------------------- | ------------------ | -------------------------------------------------------------------------------------------------------------- | | **Cassandra NTR Saturation vs Ecom Burst** | Alert band 70 / 90 | Cassandra NTR Saturation vs Ecom Burst, broken down by row. | | **Cassandra Product Table Count vs Ecom Catalog** | Merchant rule | Cassandra-distinctive XC - many merchants store catalog / inventory in Cassandra; drift = product-sync broken. | | **Cassandra Write Spike vs Ecom Order Rate** | Merchant rule | Description pending editorial review; the signal is live. | | **Dropped Mutations (24h)** | Merchant rule | DroppedMessages (MUTATION). Coordinator shed a write under load - silent data loss risk if hints also dropped. | | **Last Repair Age (hours)** | Merchant rule | Repair recency vs gc\_grace\_seconds (default 10d). Stale repair = zombie data / resurrected deletes risk. | | **Slow Reads During Checkout Window (5m)** | Merchant rule | Slow Reads During Checkout Window (5m), broken down by row. | | Signal | Alert behaviour | What it tracks | | -------------------------------------- | -------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------ | | **Coordinator Read Latency p50 (ms)** | Watch only | Description pending editorial review; the signal is live. | | **Coordinator Read Latency p95 (ms)** | Alert band 50 / 200 | ClientRequest Read latency p95 from system\_views.coordinator\_read\_latency / Read MBean. Storefront-facing. | | **Coordinator Read Latency p99 (ms)** | Alert band 100 / 500 | Description pending editorial review; the signal is live. | | **Coordinator Write Latency p99 (ms)** | Alert band 100 / 500 | ClientRequest Write latency p99. Writes are normally cheap in Cassandra - high p99 = commitlog / GC pressure. | | **Slow-Query Rate %** | Alert band 1 / 5 | Share of ClientRequest Read+Write samples over 200ms, from the EstimatedHistogram buckets behind org.apache.cassandra.metrics:type=ClientRequest,scope | | **Tombstones Scanned per Read** | Merchant rule | Cassandra-distinctive - TombstoneScannedHistogram. High tombstone reads cause latency + can trip the 100k failure threshold. | | Signal | Alert behaviour | What it tracks | | ---------------------------------------- | ------------------ | ------------------------------------------------------------------------------------------------------------ | | **Read Timeouts (24h)** | Merchant rule | ClientRequest Read Timeouts - coordinator didn't get enough replica responses inside read\_request\_timeout. | | **Request Error Rate %** | Alert band 0.1 / 1 | ClientRequest Failures + Timeouts + Unavailables as % of total reads + writes. | | **Request Error Rate Spike (>1% in 5m)** | Alert band 0.1 / 1 | Alerts for Request Error Rate Spike (>1% in 5m). | | Signal | Alert behaviour | What it tracks | | ---------------------------------------------- | ------------------ | ------------------------------------------------------------------------------------------------------ | | **Dropped Mutations or Hinted-Handoff Growth** | Merchant rule | Alerts for Dropped Mutations or Hinted-Handoff Growth. | | **Last Snapshot Age (hours)** | Alert band 24 / 72 | Most recent nodetool snapshot (or Medusa / Astra backup). Gated: requires snapshot tooling visibility. | ### Audit risks and opportunities A fix status appears only where the action, inputs, approval, verification and recovery controls are mapped. Candidate remediations are never executable. Open a check for the detail. **Severity** critical · **Outcome** Customer experience · **Fix status** Report only At 90% of the connection pool in use, the database is close to refusing new connections outright. Once it does, every part of the application that needs a fresh database connection, including new customer sessions and checkout, starts failing, not just slowing down. Vortex IQ detects and explains this; resolution is manual, with evidence and recommended steps. Reference: `DB-CAP-001` **Severity** critical · **Outcome** Run operations · **Fix status** Report only A database that runs out of disk stops accepting writes entirely, which for most stores means orders, inventory updates and customer records stop being saved, not just that the database gets slower. There is very little runway left at 90%. Vortex IQ detects and explains this; resolution is manual, with evidence and recommended steps. Reference: `DB-CAP-002` **Severity** critical · **Outcome** Run operations · **Fix status** Report only More than 1 in 100 queries is failing right now. Depending on what those queries do, this can mean orders not saving, pages failing to load product or customer data, or background jobs silently dropping work, and a rate this high in a 5-minute window is an active incident, not background noise. Vortex IQ detects and explains this; resolution is manual, with evidence and recommended steps. Reference: `DB-ERR-001` **Severity** high · **Outcome** Run operations · **Fix status** Report only If something goes wrong with this database right now, the most recent point it can be restored to is over 3 days old. Every order, customer record and inventory change since that backup would be unrecoverable in a real incident, not just delayed. Vortex IQ detects and explains this; resolution is manual, with evidence and recommended steps. Reference: `DB-BAK-001` **Severity** high · **Outcome** Run operations · **Fix status** Report only Anything reading from the replica, reports, dashboards, or read traffic split off the primary for capacity, is now up to 10+ seconds stale. If the primary fails while lag is this high, the replica is also that far behind on failover, which is a bigger problem than the staleness alone. Vortex IQ detects and explains this; resolution is manual, with evidence and recommended steps. Reference: `DB-REP-001` **Severity** high · **Outcome** Customer experience · **Fix status** Report only More than 1 in 20 queries is landing in the slow bucket. That is frequent enough to be a pattern, not noise, and it means a meaningful share of every page load or job that touches this database is paying the slow-query cost, not just an unlucky occasional request. Vortex IQ detects and explains this; resolution is manual, with evidence and recommended steps. Reference: `DB-PERF-002` **Severity** high · **Outcome** Customer experience · **Fix status** Report only One in twenty queries against this database is taking over 200ms, sustained for at least 15 minutes, not a brief spike. Any storefront page, checkout step or order sync that depends on this database inherits that slowness directly, and a sustained p95 this high is usually already visible to customer Vortex IQ detects and explains this; resolution is manual, with evidence and recommended steps. Reference: `DB-PERF-001` **Severity** medium · **Outcome** Run operations · **Fix status** Report only More than 1 in 5 reads is missing the cache and going to disk instead, which is markedly slower. This shows up as everything the database does feeling incrementally heavier, rather than as one obvious failure. Vortex IQ detects and explains this; resolution is manual, with evidence and recommended steps. Reference: `DB-CACHE-001` #### Build your own automated fixes Turn any finding into an automated fix with a Vortex IQ workflow: **over 13,000 read and write operations across more than 200 connectors** are available as building blocks, with approval, verification and rollback on every change. ### Automate approved work Vortex IQ is integrated with **10 read** and **0 write** operations across nodetool compactionstats, nodetool netstats, nodetool status, nodetool tablestats, nodetool tpstats, select \* from systemlocals on Apache Cassandra. Combine them with anything from the **over 13,000 operations across more than 200 connectors** to automate the work in your own words. Changes follow your configured approval policy: the target, proposed change, affected records, risk, reversibility and verification plan are shown before execution. [Create a workflow](https://app.vortexiq.ai/workbench/flows/create?connector=cassandra) #### Ready to build your first Apache Cassandra workflow Pick a trigger, add the operations above as steps, and every step that changes data pauses for your approval. Monitoring and audits are live now and can start any workflow you build. | Resource | Read operations | Write operations | | -------------------------------- | --------------- | ---------------- | | nodetool compactionstats | 1 | 0 | | nodetool netstats | 1 | 0 | | nodetool status | 1 | 0 | | nodetool tablestats | 1 | 0 | | nodetool tpstats | 1 | 0 | | select \* from systemlocals | 1 | 0 | | select \* from systempeers | 1 | 0 | | select \* from systemviewclients | 1 | 0 | Signed-in users see the full catalogue in the workflow builder, filtered to the sources they have connected.