Google Cloud Diagram Guide

GCP data pipeline diagram: Pub/Sub, Dataflow, BigQuery, and the renamed services

By the engineer who builds Datadef, from client work on data platforms · Reviewed August 21, 2026

The canonical Google Cloud pipeline is five boxes long, and drawing it should be quick. What slows people down is the icon set: Google has renamed a large share of its data products, the icon files still carry the old names in most packs, and searching for the mark you remember returns nothing. This page maps the current services to the current icons and draws the pipeline around them.

Try it free
7 min readFor data engineers drawing a Google Cloud pipeline with the official icon set

See it as a diagram

Everything below, as a diagram you can edit. Describe yours and see it in seconds.

281/2000• 3 credits left
Try:

No account needed · Editable canvas, not a picture

The canonical pipeline in five boxes

Pub/Sub receives events, with the topic and the subscription as separate concerns worth drawing separately when more than one consumer exists. Dataflow runs the streaming or batch job. BigQuery holds the datasets. Cloud Storage holds the landing and staging files and the Dataflow temp location. Looker Studio or Looker reads the reporting datasets.

Around that core: Cloud Composer when orchestration is a DAG, Dataproc when the processing is Spark, Datastream for change data capture out of Cloud SQL or Oracle, Data Fusion when the pipeline is built visually, Dataplex for governance and data quality, Analytics Hub when datasets are shared across organisations, and Vertex AI when a model reads the curated tables.

Draw the Pub/Sub subscription explicitly once there are two consumers. A single topic with three subscriptions is a fan-out, and it is invisible if the diagram draws one arrow out of the topic. Put the subscription type, push or pull, on the edge, because it decides who owns retries.

Half the icons people search for have been renamed

Cloud Dataflow is now Dataflow. Cloud AI Platform became Vertex AI. Data Studio became Looker Studio. Stackdriver split into Cloud Monitoring, Cloud Logging, Error Reporting, and Cloud Trace. Cloud IoT Core was retired. Cloud Dataprep, Cloud Data Fusion, and Cloud Composer dropped or kept the Cloud prefix inconsistently. Google also still publishes the pre-2025 set alongside the current one, so two diagrams built from official downloads a year apart will not match.

A mirror of a vendor set keeps whatever the vendor gave it, retired marks included, and this one is no exception: gcp-stackdriver.svg sits beside gcp-cloud-monitoring.svg, gcp-data-studio.svg beside the Looker mark, gcp-ai-platform.svg beside gcp-vertexai.svg. Searching the old name returns the old mark. That is correct behaviour for a mirror and it is the wrong thing to rely on, because nothing in the file tells you the product was renamed.

So put the burden on the label rather than the glyph. Use whichever mark your set ships and write the current product name under it. A reader recognising a slightly old icon is a much smaller problem than a reader reading "Stackdriver" on a 2026 diagram and concluding the platform predates a migration that happened years ago.

Icon terms are per vendor

Google publishes its architecture icons with its own usage conditions, and they are not the same as the AWS or Microsoft conditions. See Azure and Google Cloud icon terms before a public deck.

Draw the project and region boundaries

Google Cloud resources are scoped to projects, and a real data platform is usually three or more: an ingestion project, a data project, and one or more consumer projects, with IAM bindings at each boundary. Projects are the boundary that decides who can do what, so they are the right thing to draw as zones. Folders and the organisation node sit above them and can be a small bracket rather than another nesting layer.

Region belongs on the label, not in a corner. BigQuery datasets are regional or multi-regional, a query cannot join across locations, and a Dataflow job in the wrong region pays egress. Writing the location on each dataset and each bucket turns a class of production surprises into something a reviewer catches on the picture.

If you run a VPC Service Controls perimeter, draw it as a band around the projects inside it, with the ingress and egress rules as labels on the band. It is a control plane, not a data hop, so it does not belong in the flow.

Getting the picture without hunting for files

Describe the pipeline in one sentence and Datadef resolves each service name to its official mark and lays out the flow with labelled edges and project zones. The library holds 226 Google Cloud icons inside 2,098 files covering AWS, Azure, Kubernetes, and the data tools Google does not draw.

Write the full product name rather than the abbreviation and this goes smoothly. Abbreviations are where a cross-vendor index gets you: the alias table resolves gcs to Google Cloud Storage exactly, but there is no entry for adls, and a two-word query like "vertex ai" falls through to name similarity and can return something unrelated. The generator works from full names for that reason, and the per-service pages exist so a search engine can send you straight to one file.

Edit the canvas afterwards, then export to PNG or JPEG, or embed the live image in the repository README so the pipeline picture in the docs matches the pipeline. If you only want a file, the Google Cloud icon collection and per-service pages such as the BigQuery icon have them.

FAQ

What services belong on a GCP data pipeline diagram?

Pub/Sub for events, Dataflow for streaming or batch processing, BigQuery for the datasets, and Cloud Storage for landing and staging files, plus Cloud Composer for DAG orchestration, Datastream for change data capture, Dataplex for governance, and Looker or Looker Studio for consumption.

Which icon should I use for Vertex AI if my pack only has AI Platform?

The AI Platform mark is the predecessor of the Vertex AI mark, and most packs still carry both files side by side because a mirror keeps what the vendor published. Use whichever file your set contains, but label the node Vertex AI. The product name on the diagram matters far more to a reader than the vintage of the glyph, and nothing in the file itself will tell them the service was renamed.

Should Google Cloud projects be drawn as boundaries?

Yes. The project is the IAM and billing boundary, so it determines who can read a dataset or run a job. Drawing projects as zones and putting the cross-project bindings on the edges answers most access questions without extra text.

Do BigQuery datasets need a region on the diagram?

Yes. Datasets are regional or multi-regional, a query cannot join across locations, and a job running in the wrong region pays egress. Writing the location on each dataset and each bucket turns a common production surprise into something a reviewer catches on the picture, which is cheaper than catching it when a pipeline fails in staging.

Where does Dataplex belong on the diagram?

As a governance band across the storage and query layers rather than as a step in the flow. Dataplex organises data into lakes and zones and applies quality and cataloguing on top of existing Cloud Storage buckets and BigQuery datasets, so no data moves through it.