You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
Remove Storage Intelligence and Storage Insights resources after GCS optimizations #368
Storage Intelligence and Storage Insights were set up in the httparchive project in August 2026 for a GCS cost analysis. Once the resulting optimizations for gs://httparchive (storage class tiering, lifecycle rules, bucket location) are decided and implemented, remove the resources below. None of them are managed in Terraform.
Why
Storage Intelligence is billed at Standard edition. The 30-day trial expired on 2026-09-09 and the project's effective edition is now STANDARD. The object management fee is about $2.50 per million objects per month; with ~900M objects in gs://httparchive that is roughly $2,250/month. Check the Cloud Storage lines in billing to confirm the actual charge.
The dataset has had no new data since 2026-08-23, so the subscription currently produces nothing usable.
Findings from the Storage Intelligence data (inventory snapshot of 2026-08-22, activity 2026-08-11 to 08-23) and the bucket's current lifecycle rules. All figures are at list prices; actual billing may differ.
Current setup:results/ (~4.6 PiB) moves to Coldline at 30 days and is deleted at 400 days. crawls/ is deleted at 40 days, so there's nothing to tier there. Soft delete is off. Reads are ~0.8 TiB/month. Writes are ~1 PiB/month, coming from crawlers in 7 US regions.
Option
Saving / month
Cost
Confidence
1. Archive instead of Coldline for results/ (change the target class of the 30-day rule)
~$21.5k at rest, minus $3.6k/month in higher transition charges (Archive Class A $1.00 vs Coldline $0.40 per 10k), so **$18k net**
Nothing up front if applied only to new objects; ~$74k to also move the existing 740M objects
Medium: depends on whether results/ is read after 30 days
2. Single-region bucket for new writes
Up to ~$22k in steady state, mainly the $0.02/GiB multi-region replication fee on writes plus cheaper regional storage
~$0 if old data isn't copied and the old bucket empties itself through the existing 400-day rule. A full copy costs ~$219k (10-month payback)
Low until the crawler regions are settled
Option 1: Archive for results/
Deletion at 400 days is past Archive's 365-day minimum, which counts from object creation, so there's no early-deletion fee.
Archive reads are as fast as Coldline in GCS; the downside is a $0.05/GiB retrieval fee instead of $0.02.
Open question: is results/ read after 30 days, for example by reprocessing or backfills? Storage Intelligence object events didn't capture this bucket's reads, so the data can't answer it. If nothing reads it, this is a one-rule lifecycle change.
Option 2: single region
86% of writes come from regions other than the busiest one (us-east1). Unless the crawlers write from a single region, inter-region transfer ($17k/month) cancels most of the replication saving.
A new bucket name changes every path in the pipeline, and any external users of gs://httparchive paths.
A single region gives up multi-region redundancy.
Retention
Keeping results/ for 400 days is the biggest cost driver. Each month of retention is 0.35 PiB of Coldline ($2.5k/month). If 400 days is more than the project needs, shortening it is the simplest saving.
Suggested order
Confirm the read pattern of results/ after 30 days. If it isn't read, switch the 30-day rule to Archive.
Revisit retention.
Consider option 2 only together with a decision on where the crawlers run.
Activity
max-ostapenko commentedon Oct 29, 2026
GCS optimization options for
gs://httparchiveFindings from the Storage Intelligence data (inventory snapshot of 2026-08-22, activity 2026-08-11 to 08-23) and the bucket's current lifecycle rules. All figures are at list prices; actual billing may differ.
Current setup:
results/(~4.6 PiB) moves to Coldline at 30 days and is deleted at 400 days.crawls/is deleted at 40 days, so there's nothing to tier there. Soft delete is off. Reads are ~0.8 TiB/month. Writes are ~1 PiB/month, coming from crawlers in 7 US regions.results/(change the target class of the 30-day rule)$3.6k/month in higher transition charges (Archive Class A $1.00 vs Coldline $0.40 per 10k), so **$18k net**results/is read after 30 daysOption 1: Archive for
results/results/read after 30 days, for example by reprocessing or backfills? Storage Intelligence object events didn't capture this bucket's reads, so the data can't answer it. If nothing reads it, this is a one-rule lifecycle change.Option 2: single region
86% of writes come from regions other than the busiest one (us-east1). Unless the crawlers write from a single region, inter-region transfer ($17k/month) cancels most of the replication saving.gs://httparchivepaths.Retention
Keeping
results/for 400 days is the biggest cost driver. Each month of retention is0.35 PiB of Coldline ($2.5k/month). If 400 days is more than the project needs, shortening it is the simplest saving.Suggested order
results/after 30 days. If it isn't read, switch the 30-day rule to Archive.