Platform · Incidents & Alerting

Signals become incidents. Incidents become action.

A wall of threshold alerts is not monitoring — it is noise with a timestamp. SchemaPulse collapses related signals into one incident that carries its context, its history and its owner.

The pipeline

Signal → Alert → Correlation → Incident → Investigation → Action

  1. 01

    Signal

    A metric crosses a threshold, a replica falls behind, a cluster member leaves, a backup run finishes without verification. Raw, frequent, and on its own not worth waking anyone for.

  2. 02

    Alert

    A rule scoped to the instance and the metric decides whether this signal matters here. What is critical on a write primary can be expected on a reporting replica.

  3. 03

    Correlation

    Related alerts across members of the same cluster collapse into one incident. Correlation that ignores topology produces three incidents for one failover.

  4. 04

    Incident

    An object with severity, state, an owner, a comment thread and the full event sequence that produced it — not a notification that vanishes.

  5. 05

    Investigation

    The event history is attached, so the post-mortem is reading rather than reconstruction. Metrics, queries and topology events share the incident’s timeline.

  6. 06

    Action

    Routed to the tools your team already runs, with acknowledge and resolve available straight from the notification.

In the product

Alerts, and the incident they became

schemapulse — alertsexample data
ActiveHistoryRules
SeverityRuleScopeAgeState
CriticalBackup verification missingreporting-mssql-12d 04hOpen
WarningReplica lag > 3sbilling-rs-200h 42mAcknowledged
WarningBuffer pool hit rate < 95%orders-galera-000h 18mOpen
InfoCluster membership changeorders-galera-112h 06mResolved
MutedDisk usage > 85%analytics-pg-0Muted · maintenance

Incident · INC-2291 — correlated

02:14:08ZCluster membership change · orders-galera-1 · term 5
02:19:31ZBuffer pool hit rate fell below threshold · orders-galera-0
02:31:44ZSchema deployment recorded · orders.line_items
02:47:02ZLatency alert raised → correlated into INC-2291
02:47:04ZNotified · PagerDuty, Jira SP-418
03:02:55ZAcknowledged by on-call from notification link

Five raw signals, one incident, one page.

Fig. 1Five raw signals over 33 minutes — a membership change, a cache degradation, a schema deployment, a latency breach — collapsed into INC-2291 with its event sequence intact and the on-call acknowledgement recorded against it.

Rules

What you configure, and what it remembers

Scope
Per metric and per target — an instance, a cluster, or a project. Rules are not global thresholds pretending to fit every workload.
Severity
Assigned by the rule, so the page you get at 3am is one you agreed was worth it.
State
Open, acknowledged, muted, resolved. Every transition is recorded with who made it — a muted alert is a decision with an owner, not a gap.
Mute
Time-bounded and attributable, so silence has an expiry rather than becoming permanent by accident.
Maintenance
Scheduled windows suppress the expected, so the alerts that do fire keep their meaning.
History
Every firing, acknowledgement and mute retained, so a recurring problem is visible as a pattern rather than a series of surprises.

Correlation

Why one incident beats nine alerts

Without correlation

  • Replica lag alert — billing-rs-2
  • Membership change — orders-galera-1
  • Buffer pool hit rate — orders-galera-0
  • Query latency — digest a3f19c
  • Connection spike — orders-galera-0

Five pages. Five people looking at five fragments of one event, each deciding independently whether it matters.

With correlation

  • INC-2291 — orders-galera, elevated
  • 5 contributing signals, sequenced
  • Topology context attached
  • Owner assigned, thread open
  • Jira SP-418 linked

One page, one owner, and the reasoning already assembled — including which candidate cause the evidence actually supports.

Jira

An incident and its Jira issue stay one thing

Most integrations post a message and forget about it. The link between a SchemaPulse incident and a Jira issue is persistent and read in both directions: the incident holds the issue key, and every time you open it, SchemaPulse fetches the issue's current status,assignee and full comment thread from Jira and shows them alongside its own event history.

So the engineer working in Jira does not have to report progress back, and the DBA looking at the incident does not have to go and check. Neither system is the copy.

schemapulse — incident INC-2291example data

SchemaPulse incident

IncidentINC-2291
SeverityElevated
Scopeorders-galera
Correlated5 signals
Linked issueSP-418

Read back from Jira, shown in the incident

IssueSP-418
StatusIn Progress
AssigneePlatform on-call
Comments4, with authors
Open in Jiradeep link

Event history — both directions

02:47:02ZSPIncident opened from 5 correlated signals
02:49:10ZSPJira issue created: SP-418 — recorded as an incident event
03:02:55ZJRAssigned to Platform on-call — read back into the incident
03:14:22ZSPComment by DBA → pushed to SP-418, attributed
04:08:41ZJRStatus → In Progress — shown on the incident
09:31:07ZJRComment thread synced, 4 comments with authors
Fig. 2The incident carries the issue key; status, assignee and comments are read back from Jira on view. Comments and status changes made in SchemaPulse are pushed to Jira with their author attached.
Create from the incident
One action turns a correlated incident into a Jira issue, with its evidence and event history in the description. Creating it is itself recorded as an incident event, and a second attempt is refused rather than producing a duplicate.
Status, read back
The issue's current Jira status appears on the incident. Work moving in Jira is visible to whoever is watching the database.
Assignee, read back
Who owns it in Jira shows on the incident, so ownership is not tracked twice.
Comments, both ways
The Jira comment thread — author and text — is shown on the incident. Comments and status changes made in SchemaPulse are pushed to Jira, attributed to the person who made them.
Reopen
A recurrence can reopen the existing Jira issue rather than starting an unrelated one, keeping the history in one place.
Deep link
The incident links straight through to the issue in Jira, so nobody is searching by key.

Notifications

Fits the tools your team already runs

SchemaPulse consolidates database operations. It does not ask your engineers to watch another console alongside the one the rest of the org already lives in.

PagerDuty
Incidents raise and resolve PagerDuty alerts, and acknowledgement flows back — an incident acknowledged on a phone is acknowledged in SchemaPulse.
Jira
Push an incident to Jira as an issue with its evidence and event history attached. The link is kept, so the two stay in step.
Slack & Microsoft Teams
Posted to the channel that owns the service, with acknowledge and resolve available from the message.
Webhooks
A signed HTTP callback for your own runbook automation, status page or internal on-call system.
Email
Notification channels with test-send, and tokenised action links that work without a login.
Scoped per project
Routing follows organisation and project scoping, so a shared platform does not page the wrong team.