Time Travel With Dynamic Backfill

Steve Waterworth
May 2, 2025

Application logs provide vital information to assist with finding the root cause of an issue when an incident occurs. Under these circumstances there is no such thing as too much logging; the more information the better. However, the greater the volume of log information the greater the cost to process and store it whether using your own servers or those from one of the many SaaS providers.

Some application frameworks support changing the log level without restarting, allowing for increased detail when an issue is detected. While this is better than nothing in most circumstances the vital piece of information was logged before the issue was detected and has already been lost.

Dynamic Detail When You Need It

Grepr slips in like a shim between the log shippers and the log processing and storage servers. All logs messages are retained in low cost storage then using machine learning and a rules engine only less frequent unique messages and summaries of the more frequent noisy messages are sent through. Using these advanced techniques results in a typical reduction of 90% to the volume of messages being processed and stored by the existing logging backend.

During normal operation, this level of information provides just the right level of detail for users: low-frequency messages that are usually important like errors or misbehaviors are passed straight through, so they’re searchable and easy to find while reading through a log stream. Noisy messages that repeat often are summarized so they don’t clog the log stream and make it harder to read.

Remember no log messages are dropped, all messages are retained in low cost storage. When an incident occurs any log messages pertinent to the incident can be selectively and quickly backfilled into the logging backend. Thus providing the engineers all the detailed information they need in the tools they are familiar with.

Grepr dynamic backfill is just like going back in time to increase log level detail before the issue happened, ensuring that all the information is captured for diagnosis.

Reduce Cost Without Reducing Logging

Utilising Grepr the majority of log messages are retained in low cost storage, significantly reducing the cost of processing and storing them. The compromise between the level of detail captured in log messages and the cost now swings firmly in favour of capturing more detailed information all the time. More detail available in log messages enables engineers to diagnose issues and resolve incidents quicker, freeing them to work on new features and fixes.

Give Time Travel A Try

Give Grepr a spin and see how easy it is to start saving 90% on your logging services cost with zero interruption to your existing workflows. Stop worrying about achieving that fine balance between logging visibility and cost. With Grepr dynamic backfill you can use detailed logging at low cost and debug with your current tools.

Share this post

More blog posts

All blog posts
Product

Aggregate my log volume by 90%, yet still find anything I need? How is that possible?

Grepr uses unsupervised machine learning to reduce log volume by over 90% while preserving important data through smart, configurable aggregation. It passes low-frequency messages through unmodified, allows engineers to retain specific parameters like user IDs, and supports backfilling logs via API triggers when deeper detail is needed—such as during support tickets. For added flexibility, trace sampling can capture full logs for a subset of users, and all original logs are archived in a searchable data lake. This gives teams control, reduces noise, and enables cost-effective observability without sacrificing access to critical information.
June 30, 2025
Product

All Observability Data Is Equal But Some Is More Equal Than Others

With apologies to George Orwell. Not all Observability data is salient all the time, some data is required all the time but most data is only germane when investigating an issue.
June 24, 2025
Product

Grepr vs Vector

Vector and Grepr both function as observability data pipelines, but they differ sharply in complexity and automation. Vector, an open-source tool sponsored by Datadog, is powerful and flexible but requires extensive manual configuration, domain-specific scripting (VRL), and careful infrastructure planning. In contrast, Grepr is a fully automated, AI-driven observability platform that dynamically manages thousands of data transformations without requiring custom coding. It reduces observability costs by up to 90%, stores all data in queryable formats like Apache Iceberg on AWS S3, and integrates seamlessly with tools like Datadog and Splunk. With Grepr, organizations can deploy in minutes instead of days—without the operational overhead.
June 20, 2025

Get started free and see Grepr in action in 20 minutes.