Skip to content

Independent marketplace for verified cloud accounts

Audit Logging and Activity Monitoring From Day One

Audit logs cannot be enabled retroactively, so the window you fail to record is gone. This guide covers what to capture, where to store it, retention trade-offs and making logs searchable.

Audit logging is the only control on this list that you cannot apply retroactively. Encryption can be added later. Firewall rules can be tightened next week. But the record of what happened last Tuesday either exists or it does not, and if it does not, no amount of effort afterwards will produce it.

That asymmetry is why logging belongs in the first hour of an account’s life, before the first workload, before the network design, arguably before anything except securing the owner login.

What the record is actually for

Teams enable logging for security and then discover they use it for four different things.

The first is incident investigation: something changed, and you need to know which identity changed it and when. The second is far more common and entirely mundane: a service broke and the log shows that someone modified a configuration an hour earlier. The third is cost attribution, where the log explains who created the expensive thing. The fourth is evidence, because every compliance framework asks whether you can demonstrate what administrative actions occurred.

The second use is the one that pays for the setup. Most days the audit log is a debugging tool, and that is a good reason to make it pleasant to query rather than a dusty archive.

Turn it on before you need it

Default logging configurations are usually partial. Providers commonly record administrative actions by default and leave the more voluminous categories off, because those categories cost money to store. The default is a starting point, not a decision.

Enable logging across every region you can use, not only the ones you intend to use. Unused regions are precisely where unauthorised activity appears, because an attacker who gains access will often start somewhere nobody is watching. A log that covers only your two production regions will miss it entirely.

Do this before granting broad access to anyone else. If the log starts after the team starts working, the earliest and least reviewed period of the account’s life is unrecorded, which is generally when the most permissive configurations are created.

Control plane and data plane are different records

Providers separate two categories, and the distinction drives both your coverage and your bill.

Control plane events describe changes to the infrastructure itself: an instance created, a permission granted, a firewall rule modified, a storage bucket made public. Volume is low, value is high, and these should always be on everywhere.

Data plane events describe access to the contents: an object read, a database row queried, a message consumed. Volume can be enormous, and the cost of recording everything is real. The sensible approach is selective: enable data plane logging on the stores holding regulated or sensitive data, leave it off for build artefacts and public assets.

A third category sits alongside both: authentication events, covering every sign-in attempt, successful or not. These are low volume and worth keeping for a long time, because a pattern of failed attempts is one of the clearest early signals available. Pair this with the controls in multi-factor authentication across cloud providers, since the log is how you find out whether a factor was challenged.

Where the logs should live

A log stored in the same account it describes can be deleted by anyone who compromises that account. That is not a theoretical concern; clearing the trail is a standard step for an attacker who intends to stay.

Three measures address this. Write logs to a destination in a separate account that production identities cannot write to or delete from. Enable whatever immutability or retention lock the storage service offers, so that objects cannot be removed before a set date even by an administrator. And alert on any change to the logging configuration itself, treating a disabled trail as an incident rather than a maintenance event.

The separate account approach is one of the stronger arguments for the structure described in separate cloud accounts per environment, where a dedicated logging account exists purely as a write-only destination.

Retention, cost, and the temptation to trim

Log storage is cheap per gigabyte and expensive in aggregate, and the volume grows with your estate rather than with your revenue. The pressure to shorten retention arrives eventually.

Resist shortening the categories that matter. Intrusions are frequently discovered months after they begin, and a ninety day window often means the initial access falls outside it. A tiered approach keeps the cost sane: hold recent logs in fast, queryable storage for the period you actively investigate, then move older records to cheaper archival storage where retrieval is slower but the data still exists.

Set retention by category rather than globally. Authentication and control plane events are small and should be kept for years. High-volume data plane logs can have a shorter hot window. Where a compliance framework specifies a minimum, that minimum is a floor rather than a target, and the relevant expectations are summarised in what GDPR, HIPAA and SOC 2 require of a cloud deployment.

Making the record searchable

A log nobody can query is a storage bill with a clean conscience. The gap between having logs and being able to use them is where most of the real work sits.

Three things make the difference. Logs need to be indexed or at least queryable with a structured language, so that “every action by this identity in this hour” is one query rather than a download and a script. They need to be normalised across sources, so that application logs, infrastructure logs and authentication events share timestamps and identity fields. And they need to be in one place, because correlating across three consoles during an incident is how investigations stall.

Accurate timestamps deserve a specific mention. If clocks differ across systems, event ordering becomes unreliable and your reconstruction of an incident is a guess. Record in a single time zone, keep time synchronised, and store the raw timestamp alongside anything you display.

Write a handful of saved queries while things are calm: all actions by one identity, all permission changes in a window, all failed authentications by source, all resources created in a region you do not use. These take ten minutes each when nothing is wrong and are very hard to compose under pressure.

Alerts worth configuring on the log

Logging is passive. Monitoring is the part that tells you something without being asked. Keep the initial alert set small enough that each one gets read.

  • Any change to the logging configuration or deletion of a log destination.
  • Use of the root or owner identity for anything.
  • Creation of a new identity with administrative rights, or attachment of an administrative policy.
  • Repeated authentication failures against one identity, or successful sign-in from an unfamiliar location.
  • Resource creation in a region you do not operate in.
  • A storage container becoming publicly readable.

Each alert needs a named recipient and a documented first action. An alert that lands in a shared inbox with no owner is worse than no alert, because it manufactures the belief that someone is watching. Detection logic and the response that follows are developed further in detecting and responding to unexpected cloud activity.

Logging on an account you did not create

An inherited account needs one extra step before the configuration work: read the existing history. The log from before you arrived tells you what state the account is in and whether anything about it should concern you.

Look for identities created outside the handover, access keys issued earlier and never used since, regions with activity you cannot explain, and any gap in the record where logging was disabled. A gap is not proof of anything, but it is the one finding that justifies a more thorough review. Our guide to securing an inherited cloud account covers the surrounding checks.

Then reset the baseline: new credentials, your own logging destination, your own alert recipients. You want the record from your tenure to be unambiguously yours.

How much of this the provider gives you

Logging capability varies considerably and is worth checking before you commit. The larger platforms ship detailed audit services covering control plane, data plane and authentication, with query languages and long retention options attached. Smaller providers may offer a basic activity history in the console with limited retention and no export, which means supplementing it with your own collection at the operating system and application layer.

Neither is disqualifying, but they imply different amounts of work from you. The Google Cloud vs DigitalOcean comparison sets out that difference in operational tooling. If deep native logging is a requirement, a Buy Google Cloud Account listing gives you the full service set, while a Buy Linode Account suits teams happy to run their own collection stack on simpler infrastructure.

Related comparisons

Google Cloud vs Digital Ocean
vs

Google Cloud vs Digital Ocean

DigitalOcean optimises for how fast a developer can get something running. Google Cloud optimises for what the platform can do…

Google Cloud vs Azure
vs

Google Cloud vs Azure

Google Cloud and Microsoft Azure both cover the full hyperscaler surface, but identity, data tooling and commercial structure pull teams…

Google Cloud vs Oracle Cloud
vs

Google Cloud vs Oracle Cloud

Two enterprise platforms built around different centres of gravity. Oracle Cloud is organised around the database; Google Cloud is organised…

Google Cloud vs Kamatera
vs

Google Cloud vs Kamatera

Kamatera lets you specify a server almost component by component and talk to a human about it. Google Cloud asks…

Accounts discussed in this guide

Buy Google Cloud Account
Buy Google Cloud Account
Google Cloud

Buy Google Cloud Account

Google Cloud Platform has emerged as one of the most powerful and versatile cloud computing environments available today, and for good reason.

From$70 5 options

Keep reading

What Cloud Credits Cover, and What They Never Cover
Cost Management 7 min read

What Cloud Credits Cover, and What They Never Cover

A credit is a conditional, time limited, provider controlled entitlement, not money. What a balance actually buys, what it refuses to buy,…

Need the account, not just the theory?

Browse documented configurations across every provider in the catalog, or ask us which one fits your workload.

Telegram