S3 API Cost Analysis¶

RockLake exposes three --cost-mode presets that trade write latency for object-storage API cost. This page explains the cost model, shows the crossover points, and gives guidance on choosing a mode.

Cost model¶

Standard S3 pricing (us-east-1, as of 2024):

API class	Price per 1,000 requests
PUT/POST	$0.005
GET/HEAD	$0.0004
LIST	$0.005

RockLake's write path issues 1 PUT per SST flush, plus 1 LIST per compaction cycle. The memtable size and l0_sst_count_threshold setting together control flush frequency.

Cost-mode presets¶

`conservative`¶

l0_sst_count_threshold = 8
max_write_batch_bytes  = 64 MiB

Targets ≤ 100 PUT/hour for a 10 write-ops/s workload. At $0.005/1 000 PUTs that is ≈ $0.004/day. Best for batch-ingest pipelines that can tolerate 200–400 ms p99 write latency.

`balanced` (default)¶

l0_sst_count_threshold = 4
max_write_batch_bytes  = 32 MiB

Tuned on the TPC-H SF10 benchmark. Produces ≈ 240 PUT/hour under the benchmark load. ≈ $0.009/day. Recommended starting point for most deployments.

`latency`¶

l0_sst_count_threshold = 2
max_write_batch_bytes  = 16 MiB

Targets < 50 ms p99 write latency; issues ≈ 720 PUT/hour under load. ≈ $0.026/day on standard S3. Intended for interactive analyst workloads on S3 Express One Zone (which has lower per-request prices and higher throughput).

Monthly cost crossover vs. PostgreSQL¶

A self-managed PostgreSQL instance on a db.t3.medium RDS instance costs roughly $29/month in us-east-1. RockLake in balanced mode at 10 writes/s costs < $0.30/month** in API fees, plus storage ($0.023/GB-month for S3 Standard). Storage break-even is around **1 200 GB for a data warehouse that does not need online transaction processing.

Using `rocklake inspect api-costs`¶

rocklake inspect api-costs \
    --catalog s3://my-bucket/catalog \
    --estimate-monthly \
    --compare-postgres

Sample output:

S3 API Cost Report (last 1 h)
  PUT  count : 248
  GET  count : 1 024
  LIST count : 12
  Estimated cost (1 h)    : $0.0009
  Estimated cost (monthly): $0.65
  Equivalent Postgres (RDS db.t3.medium): ~$29.00/month
  Cost ratio: 2.2% of equivalent Postgres

Add --stream to emit one JSON object per second for real-time dashboards.

Cost-per-Operation Reference (v0.42.0)¶

The table below uses S3 Standard pricing as of 2026 (us-east-1):

Price tier	Rate
Storage	$0.023 / GB-month
PUT / COPY / POST	$0.005 / 1 000 requests
GET / SELECT / other	$0.00035 / 1 000 requests
LIST	$0.005 / 1 000 requests

S3 Express One Zone prices are approximately 50% lower for request fees and 60% lower for storage.

S3 API Calls per Catalog Operation¶

These counts are measured by the object_store request counter middleware introduced in v0.42.0, using a LocalFS backend with request interception. S3 call counts are identical to LocalFS for catalog operations because RockLake issues the same object-store operations regardless of backend.

Operation	GET	PUT	LIST	Estimated cost (S3 Standard)
`get_current_snapshot()` — warm cache	0	0	0	< $0.0000001
`get_current_snapshot()` — cold open	1–2	0	0	$0.0000007
`list_data_files(n=100)`	1	0	0	$0.00000035
`list_data_files(n=10k)`	2–4	0	0	$0.0000014
`list_data_files(n=100k)`	4–8	0	0	$0.0000028
`create_snapshot(1 file)`	0	1	0	$0.000005
`create_snapshot(100 files)`	0	1	0	$0.000005
`create_snapshot(1 000 files)`	0	1–2	0	$0.000010
`describe_table(50 cols)`	1	0	0	$0.00000035
`prune_files(100k files)`	4–8	0	0	$0.0000028
SlateDB compaction cycle	2–10	1–5	1–2	$0.000015–0.00006

Note: RockLake batches all catalog mutations for a logical commit into a single SlateDB WriteBatch, which flushes as at most 1–2 PUTs per commit (one memtable SST, one manifest update). Multi-file additions do not proportionally increase PUT count.

Monthly Cost Estimates at Typical Workloads¶

Workload	Write ops/s	Monthly S3 cost	Monthly Postgres RDS
Dev / staging	0.1	< $0.01	~$29 (db.t3.medium)
Small analytics	1	~$0.10	~$29
Medium analytics	10	~$0.65	~$87 (db.t3.large)
High-throughput ingest	100	~$6.50	~$175 (db.r6g.large)

Postgres costs are for an always-on single-AZ RDS instance. RockLake has no instance cost — only S3 API fees and storage.

Storage Cost Crossover¶

At $0.023/GB-month (S3 Standard), the storage break-even vs. a 100 GB PostgreSQL RDS db.t3.medium allocation (~$14/month storage) is ≈ 600 GB of catalog data. Almost no real-world catalog approaches 600 GB; the cost advantage is effectively unconditional.

Instrumenting Your Own Deployment¶

RockLake's object_store layer includes a request-counting wrapper (RequestCounter) that can be queried via the diagnostics API:

// From rocklake-client
let cost = catalog.api_cost_report().await?;
println!("GET: {}, PUT: {}, LIST: {}", cost.gets, cost.puts, cost.lists);

Or from the CLI:

rocklake inspect api-costs \
    --catalog s3://my-bucket/catalog \
    --estimate-monthly \
    --compare-postgres