Find the Best Cosmetic Hospitals

Compare hospitals & treatments by city — choose with confidence.

Explore Now

DataOps Lessons from an Invented Bangalore Events Desk

DataOps Lessons from an Invented Bangalore Events Desk
Tech Careers

Picture a small team that keeps a weekly guide to things happening around Bangalore: plays, gigs, stand-up nights, food pop-ups. Nobody here is real; I have invented the team so the lessons can sit inside a story rather than a lecture. What they run into, though, is exactly the kind of trouble that DataOps was coined to prevent, and I will lean on the Ops Atlas reference page for the vocabulary as we go.

A Friday that looked perfectly normal

The desk has a routine. Every night a script gathers listings from venue pages and ticketing feeds, tidies them, and publishes a table that the website reads from. On Friday afternoon the editor opens the weekend page and sees it filled. The job log says finished. Everyone relaxes.

By evening a reader has emailed to ask why no Saturday comedy shows are listed, when she knows of at least two. The editor checks and finds that Saturday has a thin set of entries. The log, still cheerful, says the script ran without errors.

What had actually happened

One ticketing feed had changed the name of a field in its output. The script did not crash; it simply read an empty value and skipped every record that lacked a date. The Atlas explains this trap with an idea from the Google SRE Workbook: each link in a pipeline may claim success even as records fall out between the links. A green status told the team that the script ran, not that the listings were right.

The painful part was timing. The page was wrong for nearly a day before a reader, not the team, noticed. In the Atlas vocabulary, the fault was detected by a consumer. It proposes tracking how often that happens as a headline figure, because it shows whether your monitoring works at all.

The Monday retrospective

Rather than blame the script, they pulled everyone round a whiteboard. The facilitator asked three questions: What did we promise? What did we check? Who found out first? The answers were uncomfortable. They had promised fresh listings each morning, but checked only that the job finished, and a stranger found the problem.

They decided on a small set of changes, modelled loosely on the first month outlined in the Atlas.

  1. Rank the pages. The weekend guide and the festival calendar became critical. The archive of old reviews became exploratory.
  2. Put the script in version control with a second pair of eyes on any change.
  3. Add basic checks: is today’s data recent, is the number of listings in the normal range for this kind of day, and does every listing carry a date and a venue?
  4. Keep an incident notebook, a plain log of what went wrong and how it was found.

The second surprise: the checks that cried wolf

Within two weeks the new volume check was firing every Monday morning. The listings on Monday were always fewer, because venues post their weekend plans late and quiet midweek nights have less to show. The team had set a single band around the average of all seven days. The Atlas has a name for this error: using limits that assume one stable process when the data really holds two, a weekday rhythm and a weekend rhythm. Its own lab exercise shows the damage: a band stretched over every day let a short load pass, while a band drawn from weekdays alone flagged that same shortfall.

The fix was easy to state: compare Mondays with Mondays, Saturdays with Saturdays. Alerts dropped to the handful that deserved a person’s attention.

Holding the new listings back

The last change was the one the team resisted longest. Rather than letting the script overwrite the live table, they wrote new data to a holding table, ran the checks there, and swapped it in only when the key ones passed. The Atlas calls this write-audit-publish, and warns that it adds delay and extra writes, so it belongs on the pages that matter most. For the weekend guide it was worth it. A bad night now meant yesterday’s page stayed up with a small note, not a half-empty one.

Cleaning up the missing Saturday

There was still the matter of repairing Friday. The team re-ran the load for that day and discovered a second lesson: a repeat run must replace the day’s records, not add to them, or the Sunday entries would have appeared twice. They rewrote the load so that each day’s partition is swapped whole. They also began comparing the count of listings per day with the count the ticketing feeds themselves report, a small reconciliation that catches anything the checks missed. None of this needed new software, only a decision to treat the listings as a product with a promise attached.

How the story ends

Three months on, the notebook shows that most problems are caught by the team before readers see them. That is the whole point of the exercise. The editors also found that their guide read better, because they had to decide what a complete listing looked like. If you enjoy the live side of the city, our features on plays, shows and venues around the city, where visitors can catch live music and the stand-up and live entertainment scene are the kind of pages a desk like this one tries to keep accurate.

What to take from it

  • A job that finishes is not proof that the output is right.
  • Choose checks that match how your data really behaves across the week.
  • Decide which pages are critical and spend effort there first.
  • Let the person who maintains the data, not the reader, be the first to know.

Further reading

The practices in the story, plus the maturity ladder and the cautions about definitions, are collected on a reference page about DataOps from the Ops Atlas. It is worth reading even if you never touch a pipeline; the habit of checking before you publish applies to any guide, itinerary or calendar.

Best Cardiac Hospitals

Find heart care options near you.

View Now