DORA - Median Time to Restore Service

What is this metric?

The time to restore service after service incidents, rollbacks, or any type of production failure happened.

Why is it important?

This metric is essential to measure the disaster control capability of your team and the robustness of the software.

Which dashboard(s) does it exist in

DORA dashboard. See live demo.

How is it calculated?

MTTR = Total incident age (in hours)/number of incidents.

If you have three incidents that happened in the given data range, one lasting 1 hour, one lasting 2 hours and one lasting 3 hours. Your MTTR will be: (1 + 2 + 3) / 3 = 2 hours.

Below are the benchmarks for different development teams from Google's report. However, it's difficult to tell which group a team falls into when the team's median time to restore service is between one week and six months. Therefore, DevLake provides its own benchmarks to address this problem:

Groups	Benchmarks	DevLake Benchmarks
Elite performers	Less than one hour	Less than one hour
High performers	Less one day	Less than one day
Medium performers	Between one day and one week	Between one day and one week
Low performers	More than six months	More than one week

Source: 2021 Accelerate State of DevOps, Google

Data Sources Required

This metric relies on:

• Deployments collected in one of the following ways:
  • Open APIs of Jenkins, GitLab, GitHub, etc.
  • Webhook for general CI tools.
  • Releases and PR/MRs from GitHub, GitLab APIs, etc.
• Incidents collected in one of the following ways:
  • Issue tracking tools such as Jira, TAPD, GitHub, etc.
  • Incident or Service Monitoring tools such as PagerDuty, ServiceNow, etc.

Transformation Rules Required

This metric relies on:

• Deployment configuration in Jenkins, GitLab or GitHub transformation rules to let DevLake know what CI builds/jobs can be regarded as Deployments.
• Incident configuration in Jira, GitHub or TAPD transformation rules to let DevLake know what CI builds/jobs can be regarded as Incidents.

SQL Queries

If you want to measure the monthly trend of median time to restore service as the picture shown below, run the following SQL in Grafana.

with _incidents as (
-- get the incident count each month
    SELECT
        date_format(created_date,'%y/%m') as month,
        cast(lead_time_minutes as signed) as lead_time_minutes
    FROM
        issues
    WHERE
        type = 'INCIDENT'
),

_find_median_mttr_each_month as (
    SELECT 
        x.*
    from _incidents x join _incidents y on x.month = y.month
    WHERE x.lead_time_minutes is not null and y.lead_time_minutes is not null
    GROUP BY x.month, x.lead_time_minutes
    HAVING SUM(SIGN(1-SIGN(y.lead_time_minutes-x.lead_time_minutes)))/COUNT(*) > 0.5
),

_find_mttr_rank_each_month as (
    SELECT
        *,
        rank() over(PARTITION BY month ORDER BY lead_time_minutes) as _rank 
    FROM
        _find_median_mttr_each_month
),

_mttr as (
    SELECT
        month,
        lead_time_minutes as med_time_to_resolve
    from _find_mttr_rank_each_month
    WHERE _rank = 1
),

_calendar_months as(
-- deal with the month with no incidents
    SELECT date_format(CAST((SYSDATE()-INTERVAL (month_index) MONTH) AS date), '%y/%m') as month
    FROM ( SELECT 0 month_index
            UNION ALL SELECT   1  UNION ALL SELECT   2 UNION ALL SELECT   3
            UNION ALL SELECT   4  UNION ALL SELECT   5 UNION ALL SELECT   6
            UNION ALL SELECT   7  UNION ALL SELECT   8 UNION ALL SELECT   9
            UNION ALL SELECT   10 UNION ALL SELECT  11
        ) month_index
    WHERE (SYSDATE()-INTERVAL (month_index) MONTH) > SYSDATE()-INTERVAL 6 MONTH 
)

SELECT 
    cm.month,
    case 
        when m.med_time_to_resolve is null then 0 
        else m.med_time_to_resolve/60 end as med_time_to_resolve_in_hour
FROM 
    _calendar_months cm
    left join _mttr m on cm.month = m.month
ORDER BY 1

If you want to measure in which category your team falls into as the picture shown below, run the following SQL in Grafana.

with _incidents as (
-- get the incidents created within the selected time period in the top-right corner
    SELECT
        cast(lead_time_minutes as signed) as lead_time_minutes
    FROM
        issues
    WHERE
        type = 'INCIDENT'
        and $__timeFilter(created_date)
),

_median_mttr as (
    SELECT 
        x.lead_time_minutes as med_time_to_resolve
    from _incidents x, _incidents y
    WHERE x.lead_time_minutes is not null and y.lead_time_minutes is not null
    GROUP BY x.lead_time_minutes
    HAVING SUM(SIGN(1-SIGN(y.lead_time_minutes-x.lead_time_minutes)))/COUNT(*) > 0.5
    LIMIT 1
)

SELECT 
    case
        WHEN med_time_to_resolve < 60  then "Less than one hour"
    WHEN med_time_to_resolve < 24 * 60 then "Less than one Day"
    WHEN med_time_to_resolve < 7 * 24 * 60  then "Between one day and one week"
    ELSE "More than one week"
    END as med_time_to_resolve
FROM 
    _median_mttr

How to improve?

• Use automated tools to quickly report failure
• Prioritize recovery when a failure happens
• Establish a go-to action plan to respond to failures immediately
• Reduce the deployment time for failure-fixing