Module blob_read 

Concurrent S3Blob reads, plus a steady stream of PUTs through the same client, against an S3-compatible store.

Written to measure GCS read throughput (PER-136). Over HTTP/2, the SDK’s HTTP client sends every request to a host over one connection, and one connection to GCS carried only 300 to 400 MiB/s. mz_aws_util::pooled_http_client spreads requests over more connections. Compare --client sdk with --client pooled to recheck that, or to tune the pool’s size.

Structs

BlobReadArgs

Concurrent persist blob reads against an S3-compatible store.

HeaderWaits 🔒

Per HTTP method, the time from each request reaching the HTTP client until its response head arrives or it fails. Includes waiting for a connection, or for a free stream on a shared HTTP/2 connection. Each SDK retry is a separate request.

PutStats 🔒

ReadArgs 🔒

ReaderStats 🔒

StoreArgs 🔒

TimedConnector 🔒

TimedHttpClient 🔒

Wraps an HttpClient to record HeaderWaits.

UploadArgs 🔒

Enums

ClientKind 🔒

Command 🔒

Constants

PUT_KEYS 🔒

PUTs cycle through this many keys under PUT_KEY_PREFIX, overwriting old ones, so they don’t pile up in the bucket.

PUT_KEY_PREFIX 🔒

Keeps PUT objects out of the keys that readers fetch.

Functions

open 🔒

Opens an S3Blob with persist’s production timeouts and no hedging.

read 🔒

read_timeouts 🔒

The SDK’s read timeout firings so far: the mz_persist_s3_read_timeouts counter, which can’t be read from S3BlobMetrics outside its crate.

run 🔒

Runs the given blob-read bench command.

summary 🔒

Count, p50, p99 and max of samples.

upload 🔒