Online directories of advocates and law firms depend on a lot of quiet automation. A visitor submits an enquiry form, a confirmation goes out, the enquiry is routed to the right profile owner, a reminder is sent if nobody replies, and a summary lands in an inbox the next morning. Behind these steps, many platforms use small cloud functions. People who run such platforms, or who simply wonder why an enquiry email arrived twice, often ask questions about how that machinery works. Here are answers, in plain language, built around the discipline known as ServerlessOps. What follows is general information on how software behaves; nothing here is legal guidance, and no particular advocate or firm is being discussed.
Q: What does ServerlessOps actually mean?
Think of it as the upkeep of small event-driven functions: the way they get triggered, what follows a failed attempt, how repeated deliveries are absorbed, the limit on parallel work, the occasional slow start and the price of each run. The Ops Atlas, maintained by xOps School, explains in its entry on the discipline that the label was most used from about 2016 to 2020, that no standard or certification defines it and that the practice itself has become common.
Q: Why would an enquiry be sent twice?
Because the systems that carry messages generally promise “at least once” delivery. If an acknowledgement is slow, the sender tries again. The Atlas warns against trusting any platform to guarantee exactly-once handling, pointing out, for instance, that a first-in-first-out queue product removes duplicates only inside a five-minute window. The fix on the platform side is to give each enquiry a stable reference and to check it before sending anything, which engineers call idempotency.
Q: Which reference should the platform use to spot a repeat?
One that belongs to the business, such as an enquiry number, not one that belongs to the plumbing. The entry explains that when messages are replayed from a holding queue they receive new internal identifiers, so a check built on those identifiers will treat a replay as brand new. That is how duplicate emails sneak past an otherwise careful setup.
Q: What happens to an enquiry that fails to send?
In a well-run system, it lands in a holding queue. The Atlas lists conditions: someone responsible for the queue, an alert when anything arrives and a shelf life that exceeds the main queue’s. Its example is fourteen days against four. If the platform has none of these, a failed enquiry may simply disappear, which for a directory means a visitor who never hears back.
The entry also explains that for background invocations the platform keeps events for a limited time (up to six hours on its main example) and retries a couple of times, so a failure destination is what turns “dropped” into “kept for review”.
Q: Is a long-running process ever a problem?
It can be. When one function calls another and waits, the caller is billed for waiting. The Atlas lists synchronous function chains among its anti-patterns. A better approach for directories is to hand off slow work, such as generating a profile PDF or sending an email digest, to a background queue and let the visitor’s page return immediately.
Q: How do timeouts fit together?
They sit inside one another. An email-service call should give up ahead of the function, while the window during which a queue hides an in-flight message needs to be six times the function limit or longer. The platform’s maximum for a standard function is 900 seconds, though directory tasks usually need seconds. Another detail from the entry: a different event service drops events after 24 hours or 185 delivery attempts, so you cannot rely on endless patience from the platform.
Q: Can a busy day take the site down?
It can if one part scales faster than another. A burst of enquiries after a media mention could start hundreds of functions at once, each hitting the same database. The Atlas recommends a ceiling for concurrency, with peak use kept under seventy percent of the cap and a warning when headroom falls below thirty percent. The regional default on its main platform is 1,000. Its lab shows how a slow partner can push parallel runs from 84 to 840 against that quota, simply because each run waits longer.
Q: Does a pay-per-use model keep the bill low?
Often, but not always. For a directory whose traffic comes in waves, per-use pricing suits well. The Atlas warns, though, that “always cheaper” is a myth: steady heavy workloads may fit containers better. It also notes that billing counts the time a function spends waiting and that initialisation code is included in billed duration. One open-source tool it mentions, a power-tuning utility, helps to find the memory setting that balances speed against cost most sensibly, and it makes real calls to the function, so it is safe to run only when those calls cause no side effects.
Q: Should we keep functions warm with a timer?
Probably not. The entry says slow first starts affect under one percent of invocations and that scheduled pings are not the answer. If an enquiry form really needs an instant response, there are platform features for that purpose.
Q: Who should own all this?
Every automated step should have a named owner, even on a small team. Ownership sounds bureaucratic, but it simply means that when an alert fires there is one person who knows it is theirs. For a directory, that might be the technical lead for the enquiry form, the operations coordinator for reminders and an external developer for the digest job. Write the names next to the steps in a shared sheet, and review it each quarter so that people who have left the team are replaced before an alert lands in an abandoned inbox.
Q: What should a platform owner measure?
The Atlas suggests outcome measures: the share of enquiries completed end to end, how often a message was sent twice, how many items are in the failure queue and how old the oldest is, how stale the backlog is, how often the platform turned work away at a limit, how much spare room is left, the share of slow first starts and the cost per million events. A simple monthly one-page review of those numbers is more valuable than a wall of dashboards.
Q: How do I know if our setup is mature?
The entry proposes five rungs: ad hoc, defined, safe, measured and optimised. It describes the third level as being reached once the team has replayed a flood of duplicates in a rehearsal and confirmed that nothing happened twice. The entry’s ten-thousand-order simulation shows how wide the gap can be: an unprotected handler repeated 504 charges, while one keyed on a business reference repeated none.
Q: Where can I read more about how directories work for visitors?
Our location-based guides illustrate what visitors see on the public side of a platform like this. For instance, the city page for Guwahati’s listed IP advocates and the sibling page for the same practice area in Madurai are examples of pages that rely on well-behaved enquiry handling underneath. Readers should consult an advocate directly for any legal question; these pages and this article are informational.
Q: What is the first thing to do this month?
Follow the Atlas outline for the first thirty days: list every trigger, every holding queue and every timeout. Then ask of each step, what would happen if this ran twice? In the following month, protect the five steps that send messages or change records, add partial-failure handling and a cap, and test with a deliberate flood of duplicates. By the third month, review cost per event, hold a game day and add automatic policy checks to your release process.
A big team is not a prerequisite. Curiosity, a short list of questions and patience with a notebook are enough. A directory that answers those questions is one where an enquiry, once submitted, arrives once, in the right place and on time.
Related posts:

Legal Deadlines for Beginners: A Primer and Mini-Glossary

Preparing to Meet a Lawyer? Start With a Simple Weekly Review

How Open Process Information Is Changing Legal Help in India

Why Law Firm Tech Hires Fail, and the Verification Fix

Why Law Office Replies Go Quiet, and a Rota That Fixes It
