Skip to content

Running an Incident Postmortem

Incidents & Maintenance 13 min read
On this page

After an outage you have to write up what happened, and often prove it to a customer, a hosting provider or an auditor. Uptimia holds most of the evidence already:

  • the incident record, with the probes that detected it
  • the individual check that failed
  • the alerting timeline, with who was paged and when someone acknowledged
  • the note you left on the monitor

Collect them in the order below. If you only need to know which button produces CSV and which produces a PDF, read Exporting Your Data instead.

Before You Start #

  • Every role that can see monitors can open an incident and download its PDF or HTML. Create public link, Revoke, Delete incident, Open manual incident and Post status update need Owner, Admin or Editor. Setting Availability impact to Doesn't count or Count part of it needs one of those roles and the Professional plan or above.
  • Your plan sets how far back you can read: Free 7 days, Basic 90 days, Professional 365 days, Enterprise 730 days, Trial 30 days. See Data and Log Retention.
  • Acknowledgment, MTTA and grouped-incident timelines exist only for monitors with an escalation policy attached. Creating a policy requires the Professional plan or above. On Free and Basic an incident still records the alerts it sent, but there is no ladder to narrate.

What Uptimia Can and Cannot Tell You #

Uptimia records what a probe saw from outside your infrastructure. It cannot read your server logs, stack traces or database state, so the postmortem's "why" still comes from your side. Uptimia settles three things: the when, the from where, and the who was told.

Seven boundaries decide whether the evidence you need still exists:

BoundaryWhat it means for a postmortem
Maintenance, paused and suspended monitorsResults that arrive while a monitor is in a maintenance window, paused or suspended are dropped on arrival: no log row, no incident, no alert. The period is a gap in the record. Nothing was logged there, passing or failing.
Real User monitorsProduce no check rows at all, so there is nothing to open, export or share.
Server monitorsAgent samples are kept for 24 hours. Open an older server check and the drawer says "This check's metrics are no longer available — agent samples are kept for 24 hours." Pull server evidence the same day.
Deleted incidentsDeleting an incident is permanent, and the Delete incident item in the More actions menu works only once the incident has ended. Any status page thread linked to the incident is deleted with it, and its public links stop working. Export before you delete.
Your readable windowOn a monitor's own history a request for a longer range is silently raised to your plan's floor. Upgrading widens that window on the next request, with no backfill and no waiting. It cannot bring back rows the nightly cleanup has already deleted. Uptime check logs are the one family where an upgrade can reach less far back than the new window. See Data and Log Retention.
Manual incidentsAn incident someone opened by hand with Open manual incident has no failure evidence: there is no Errors seen list, and the location card, titled Latest check from each location (Checks from each location once it is resolved), shows each probe's own result rather than where it failed. Its headline is the message that person typed, and the line under it reads "Opened by <who> · since <time>". While it is open, Uptimia opens no automatic incident on that monitor, but checks and logs keep running, so the check log still shows what the probes saw. If the manual incident replaced an automatic one, the automatic incident's timeline ends with "Closed by a manual incident, no recovery alert was sent."
Availability impactIf someone set an incident to Doesn't count or Count part of it, the incident, its alerts, its logs and its recorded duration stay as they were. Only the downtime used for uptime figures, reports and public status pages changes. The incident page then shows an Availability impact card with the outcome and the reason given.
The Availability impact dialog on an ended incident, with Count part of it selected, the Count from and Count until times, a reason typed in, and the counted downtime it leaves.Enlarge

Reconstruct the Timeline #

  1. Go to Incidents → History and set the date range that covers the outage.
  2. Click the incident. It opens as its own page, unlike the check detail drawer. If the incident was part of an alert group you are redirected to the group page, which carries the richer timeline.
  3. Read the summary card at the top: the error as the headline with its severity, then Started, Ended, the duration (Down for on an outage, Slow for on a slow Speed incident) and Alerts. A group page adds Acknowledged.

Below the summary sit the evidence cards. Which ones appear depends on the monitor type:

  • Where it is failing (Where it failed once the incident has ended): one row per probe location, with its IP address, the time of the check, the Result and the Response. This is the table a hosting provider or CDN will ask for, because it shows the failure was seen from several networks at once.
  • What the monitor saw: Errors seen with how many times each occurred, the Screenshot at failure, and collapsible Response headers, Response body and Traceroute blocks.
  • Family cards such as Steps and Step screenshots for Transaction, Load time by location for Speed, Certificate for SSL, What changed for DNS and What we found for Security.

On a group page the Monitors in this group card lists every member with its error and when it went down, and the evidence cards show one member at a time: pick the member in the evidence switcher.

A single incident page showing the summary card with the error and duration, the 'Where it is failing' table of locations, and the note and alert timeline in the right rail.Enlarge

Every timestamp on the incident page and in an incident or check export uses your account's time zone. The timeline card names it at the bottom, for example "Times in America/New_York". State the zone in your writeup when the reader is somewhere else.

For the individual checks either side of the failure, go to Logs, filter to the monitor and the time range, and keep the All checks tab selected. That gets you the passing check before the outage, the first failing one and the recovery. Failed only lists the failures alone. Monitoring Logs covers that surface; Incidents covers the incident record itself.

Read the Alerting Side #

The Timeline card records the response. On a single incident it lists which contact received which alert and when. On a grouped incident it runs through the whole ladder, oldest first:

  • Down digest sent, tagged Step 1, with the contacts it reached. On a group that holds only one monitor the same row reads Down alert sent.
  • A joined row for each monitor that failed inside the join window, either riding the digest or announced in a batched Join update sent.
  • Escalation alert sent, tagged Step 2, Step 3 and onward, or Repeat 1 for a repeat round, each listing the contacts that rung reached.
  • Acknowledged by: the responder and the source, with a sub-line naming the step the acknowledgment paused.
  • Partial recovery: recorded, with no alert sent.
  • The final recovery row, with the one recovery alert that went to everyone who had been alerted.

While an incident is still open, steps that have not fired yet appear as dashed rows reading "Step 2 — fires 09:56 · in 12 min". That is how you tell a rung that fired from one that was only armed. After an acknowledgment, the step that was due reads Step 3 — on hold instead.

The Acknowledged fact on a group page answers how fast your team reacted. It names who took the incident, and the line under it gives the clock time and the MTTA: the whole minutes from the moment the incident was confirmed to the acknowledgment, rounded down. A further line names the source, such as "via control panel" when someone clicked Acknowledge in Uptimia. A single incident on a monitor with no escalation policy has no group page, so there is no acknowledgment and no MTTA to report. The mechanics are in Acknowledging Incidents and MTTA.

A grouped incident page for 3 of 5 monitors down, marked Acknowledged, with the Acknowledged cell naming Alex Rivera, MTTA 6 min and escalation paused, and a timeline showing the down digest, the join update, the step 2 alert, the acknowledgement, two partial recoveries and step 3 on hold because of the acknowledgement.Enlarge

Pull One Failing Check as Evidence #

A hosting provider or CDN will often accept a single check as the artifact.

  1. Open the check from Logs, or from the Check log card on the monitor's page. The check detail drawer opens.
  2. Click Export.
  3. Choose Download PDF, Download HTML or Create Public Link.
The check detail slide-over open on a failed domain check with its Export dropdown expanded, showing Download PDF, Download HTML and Create Public Link.Enlarge

The file downloads as Monitoring-Log.pdf or Monitoring-Log.html, a standalone snapshot of that one check. It carries:

  • the status verdict and error
  • the Check details group: the URL, protocol, checked-at time, response time and the probe that ran it
  • the family's own facts, such as certificate issuer and validity for SSL, registration dates for Domain, and signal and lateness for Heartbeat
  • the response captured at that check

Eleven monitor families can be exported this way. Real User is the exception: it produces no check rows, so there is no check to open and no export to take.

Export the Incident and the Filtered History #

For the incident itself, use the Export menu in the page header: Download PDF, Download HTML, Download alert deliveries (CSV), and Create public link for Owner, Admin and Editor seats. The files are named Root-Cause-Analysis.pdf or Root-Cause-Analysis.html.

The Export menu open in an incident's header, showing Download PDF, Download HTML, Download alert deliveries (CSV) and Create public link next to Open monitor and Post status update.Enlarge

Uptimia rebuilds the whole incident on the server: every member monitor, its root-cause evidence, and the full event timeline including the escalation rungs. The footer is stamped "This is a point-in-time snapshot of incident #N", followed by the capture time and the time zone it is printed in. The PDF version opens every collapsible evidence block, so no evidence stays hidden.

For the incident set rather than one incident, go to Incidents → History and click Export CSV. The download follows the date range and filters you have applied. There is no CSV button on the Active view: the export is history-based. Besides the times, duration and alert counts, each row carries Opened by and Closed by manual incident for manual incidents, and Availability impact, Counted seconds and Impact reason. Duration stays the recorded length, so use Counted seconds when you report downtime. Exporting Your Data lists all the columns.

Add Context That Survives #

A note on the monitor is the one piece of postmortem context you can write before the next incident. Set it in the monitor editor on the Advanced tab under Notes, up to 500 characters. Server monitors keep it in the Server Name card. The note then appears in four places: the Your note card on the incident page (on a group page it names the monitor, as in Your note · Checkout API), alert emails, and webhook payloads. Write it for the responder at 03:00: which provider to call, which backup job to check first, the runbook URL.

An incident page with the monitor's note in the 'Your note' card at the top of the right rail, carrying the runbook line the responder needs at 3am.Enlarge

Note: The note does not travel into the PDF, HTML or public-link export. It never reaches status pages, Slack, SMS or any channel other than email and webhooks. Full coverage is in Monitor Notes.

Share With People Who Have No Account #

Create public link saves a static HTML copy of the document at its own web address. Its filename carries a random 32-character hash. Anyone holding that URL can read it: there is no sign-in, no password and no expiry date.

An incident's links can be withdrawn. Under the incident page header, Owner, Admin and Editor seats see each working link with Copy and Revoke. After you confirm Revoke, anyone who opens the link gets a page-not-found error. This cannot be undone, but you can create a new link afterwards. Deleting the incident also stops its links. Public links of a single check cannot be revoked. Neither can some older incident links that are not listed under the header, and those also survive deleting the incident, so treat them as permanent.

An incident page with its public link listed under the header and the Revoke public link confirmation open, warning that anyone who opens the link will get a page-not-found error.Enlarge

Availability impact changes what a public link can show:

  • Doesn't count: Create public link is not offered, and existing links stop working.
  • Count part of it: existing links stop working. A new link shows only the counted stretch, with start and end times and no evidence, locations or alert events.
  • Setting the incident back to Counts in full does not bring old links back. Create a new one.

Know what you are handing over:

  • A check export redacts one thing. A Heartbeat monitor's ping URL is masked to its last four characters, because that token is the monitor's only credential. Everything else appears in full: the monitored URL, the probe, and the response captured at that check.
  • An incident public link names nobody. Recipient rows read "Recipient" with the channel only, and the page leaves out who opened, resolved or acknowledged the incident, the escalation policy name and the Alert delivery table. A manual incident's message stays.
  • A downloaded incident PDF or HTML redacts nothing. The timeline's recipient boxes name each contact and print their email address or phone number, up to twelve per box, and the file ends with an Alert delivery table of every recipient, channel and status.

Warning: before attaching a downloaded incident PDF or HTML to a message outside your team, check the recipient boxes and the Alert delivery table. If they carry contacts you do not want to disclose, send the incident public link instead, or the check-level export.

Close the Loop #

If the incident was visible to customers, post the closing update on your status page. The Post status update button in the incident page header pre-fills a thread from the incident. It works on Uptime, Speed and Transaction incidents only. Elsewhere the button is grayed out, with the hint "Status updates are available for uptime, speed and transaction incidents." On a manual incident the draft starts from the message that was typed when it was opened.

If the work that follows is planned, schedule a maintenance window for it. Monitors leave the check queue for the duration, so the remediation does not become the next incident in your history.

A Postmortem Built From These Exports #

Section of your writeupThe artifact that proves it
Detection time and durationIncident summary card: Started, Ended and Down for
Blast radiusGrouped incident: the member list under Monitors in this group. The Monitors down counter shows only while the incident is still open.
Where it was visible fromThe Where it failed table: location, IP and per-probe response
What the endpoint returnedWhat the monitor saw: Errors seen plus the response blocks, or one check export
Response speedAcknowledged and MTTA, plus the fired rungs in the timeline
Who was paged, and who was notTimeline recipient boxes on each fired step
Downtime you reportCounted seconds in the history CSV, and the Availability impact card with its reason when an override is set
Context the responder hadThe Your note card
The record for the ticketRoot-Cause-Analysis.pdf for the incident, Monitoring-Log.pdf for the single check, and the history CSV for the pattern

Was this helpful?

Your answer helps the team improve this article.

Still need help?

Uptimia Bot answers right away and passes you to the team when it can't.

Source: https://help.uptimia.com/help/a/incident-postmortem-with-uptimia