Consider change the estimated byte size for file based sources for better initial autoscaling (on runners which support it; currently only Dataflow). Currently the estimate is based on total size, for compressed file we could add a factor multiplier (~5) for compressed files in a blob go produce a better estimate.
Imported from Jira BEAM-8495. Original Jira may contain additional context.
Reported by: altay.