* Add multiparting to s3client.
Fix boost::urls::parse_uri 's dislike of credentialed blobstore urls.
* fdbclient/BulkLoading.cpp
Add blobstore regex to extract credentials before feeding the boost
parse_uri.
* fdbclient/include/fdbclient/S3BlobStore.h
* fdbclient/S3BlobStore.actor.cpp
Add cleanup of failed multipart -- abortMultiPartUpload l(s3 will do
this in the background eventually but lets clean up after ourselves).
Also add getObjectRangeMD5 so can do multipart checksumming.
* fdbclient/S3Client.actor.cpp
Change upload file and download file to do multipart always.
Retry too.
* fdbclient/S3Client_cli.actor.cpp
Add command line to trace rather than output.
* Address Zhe review
* More logging around part upload and download
* Undo assert that proved incorrect; restore the old length math
doing copy in readObject.
Cleanup around TraceEvents in HTTTP.actor.
* Undo commented out cleanup -- for debugging
* formatting
---------
Co-authored-by: stack <stack@duboce.com>
We learnt that a new connection needs to be made for each HTTP
request through proxy to AWS S3, thus it would fail when try
to re-use the connections and see retryable errors.
Meanwhile, delay between each retry grow 2x each time it failed,
if the connection pool has a larger size, the delay can be long.
As a result, this change adds Knobs to cap the max delay of
retryable errors, including one for general retryable errors,
and the other only for connection failure error.
This change also adds more logging for debugging.
* EaR: REST based Simulated KMS Vault request hanlder interface
Description
diff-1: Address review comments
Improve unit test case coverage
diff-2: Extend RESTKmsConnectorUtil to generate HTTP::Header
EaR simulation testing is currently driven using SimKmsConnector
interface, it exposes endpoints directly invoked by EKP to fetch
encryption keys. Approach avoids testing RESTKms communication
path. Recently FDB codebase got extended by adding HTTPServer
interface, which was a gap prohibiting end-to-end testing of
EaR code.
Patch proposes following changes:
1. Refactor RESTKmsConnector to move common code and definitions
to RESTKmsConnectorUtil namespace
2. Introduce RESTSimKmsVault accepting HTTP format requests and
providing appropriate HTTP response.
Testing
RESTUnit 100K + 5k valgrind
devRunCorrectness 100K
Testing
* Passes existing tests
* adding http unit test for wrong md5 sum
* Added new HTTPKeyValueStore workload to test long-running http clients
* fixing warnings
1. For v4 signature, http request path should be encoded, currently only parameters are encoded.
2. For v2 signature, http request date header should be in form of either the http date header or x-amz-date, currently in form of ISO 8601
3. Empty response header value cause http_bad_response error.
4. Skip verifying MD5 for range get request and if http response code 206 is returned.
Michael Stack