Datadog Interview Questions — Page 1

Datadog·Software Engineer·Technical Phone Screen

Apr 2026

Phone screen for a software engineer role at Datadog. Just one problem, the classic log and query thing.

Design a system that supports writing log entries and querying them efficiently.

“Pretty much the whole interview.”

Datadog·Software Engineer·Technical Phone Screen

Apr 2026

Datadog SWE interview with a pretty gnarly streaming/matching problem. The kind of question where the core idea isn't too bad but the edge cases and efficiency requirements will eat you alive if you're not careful.

You're given a stream of strings, each prefixed with either 'Q:' (a query, which is a set of space-separated words) or 'L:' (a log line). Queries should be indexed as they arrive. For each log line received, output all previously seen queries where every word in that query appears at least once in the log line. Matching is case-sensitive and word boundaries are defined by spaces. How do you design and implement this efficiently for online processing?

“My first instinct was to just iterate over all stored queries for each log and check word membership, which works but feels bad at scale.”

View Post

Datadog·Machine Learning Engineer·Technical Phone Screen

Apr 2026

Datadog ML engineer interview with a practical API coding task. They gave you a real integration problem and AI tools were fair game, which was a nice change from the usual whiteboard stuff.

Connect to the Snowflake API, submit a query, and retrieve the query status.
Follow-up: what would you do if the query returns too much data?

“AI tools were allowed so I leaned on them pretty hard for the auth setup.”

View Post

Datadog·Machine Learning Engineer·Technical Phone Screen

Apr 2026

ML engineer round at Datadog that was basically one big deep-dive into loss functions for imbalanced classification. The question sounds contained but the follow-ups kept branching and I wasn't ready for how far they wanted to go.

Implement focal loss from scratch for binary classification using only basic tensor operations. Your implementation must handle raw logits (not probabilities), support a focusing parameter gamma and a class-balancing weight alpha, be numerically stable for large-magnitude inputs, and support none/mean/sum reduction modes.
How would you extend this to multi-class focal loss using softmax over C classes? How does p_t change, and how do you efficiently index the true class probability?
Why does the focusing term help even when you already have alpha-balancing? Concretely, how does gamma change the gradient contribution of an easy example with p_t around 0.9 versus a hard one with p_t around 0.1?
How do you choose alpha and gamma in practice and how are they coupled? What happens to the loss behavior as gamma approaches infinity?
Some implementations stop the gradient through the modulating factor (1 - p_t)^gamma. What is the motivation for doing that, and what trade-off does it introduce?

“My first instinct was to just do sigmoid then log, which is exactly the wrong move.”

View Post

Datadog·Software Engineer·Technical Phone Screen

Apr 2026

Datadog software engineering interview with a coding problem that doubles as a mini system design exercise. The core question was deceptively tricky once the follow-ups kicked in.

Write a function that reads a mixed stream of queries (prefixed 'Q:') and logs (prefixed 'L:'), assigns each new query an incrementing ID, prints an acknowledgment per query, and annotates each log line with the IDs of any queries it matches.
How would you redesign this to handle extremely high data volumes efficiently?
How would you add support for deleting queries, and what inefficiencies does that introduce in your current design?

“Started okay, got the basic parsing working pretty fast.”

View Post

Datadog·Software Engineer·Onsite - Product Sense / Strategy

Apr 2026

Bizops interview at Datadog with a classic go-to-market case question about sales territory design. Short and to the point, but the kind of question that exposes whether you actually think in frameworks or just wing it.

How would you distribute sales reps across the country?

“I went straight to market size and density, which felt right, but I skipped over rep capacity and cost-to-serve for a while and had to backtrack.”

View Post

Datadog·Machine Learning Engineer·Onsite - System Design / Architecture

May 2026

Datadog ML engineer interview that went deep into monitoring system design, specifically around alerting pipelines for web server latency. The question was meaty and I don't think I fully nailed the on-call policy portion, but the technical detection side felt okay.

Design a pipeline to monitor latency across dozens of web servers (each reporting average response time per second), including algorithms to detect degraded performance and a paging policy for on-call engineers that handles ack, snooze, and re-page intervals to reduce alert fatigue without missing real regressions.

“This is basically two questions stitched together and I treated them separately which I think hurt me.”

View Post

Datadog·Software Engineer·Technical Phone Screen

May 2026

Datadog SWE interview with a low-level systems design question about buffered I/O. The problem was more involved than it looked and had a lot of moving parts to keep track of.

Design and implement a BufferedWriter class that wraps a constrained device API. The API accepts at most M bytes per call and has high per-call overhead. Your class needs write(), flush(), and close() methods, must preserve byte order across calls, minimize total calls to the device, handle inputs larger than the internal buffer, and guarantee no data loss. Walk through your buffer management strategy, tricky edge cases between consecutive writes, time/space complexity, and optionally how you'd make it thread-safe.

“I liked this question more than I expected to.”

View Post

Datadog·Software Engineer·Technical Phone Screen

May 2026

Datadog coding round, one problem the whole session. It's a tree reconstruction question dressed up in observability language which I thought was a nice touch given the domain, but the core of it is just building a tree from a flat list of nodes.

Given a flat list of span objects (each with a span ID, parent ID, start time, and possibly other fields), reconstruct the full tree structure. A span is a root if its parent ID is null or points to an ID that doesn't exist in the input. Children at every level must be sorted by start time. Return the list of root spans, each with a nested children field.

“The observability framing made me overthink it at first.”

View Post

Datadog·Software Engineer·Technical Phone Screen

May 2026

Datadog SWE interview with a log matching problem that starts simple and gets interesting fast. The follow-up is where the real interview happens.

Given a stream of log records and a collection of queries (each query being a set of words), how would you determine which logs match which queries? A log matches a query if every word in the query appears in the log.

“The naive approach is obvious: for every log-query pair, scan the log for each query word.”

View Post