This dialer supports source-based routing and is already used in cache.
Change-Id: Ic7852edd2faea4e5d8667221e6f681cc82bb143a
Signed-off-by: Evgenii Stratonikov <e.stratonikov@yadro.com>
There are two problems with the current approach:
1. For TLS connections we need different transport credentials.
2. grpc.NewClient() considers scheme from `URIAddr()` as a scheme for a
resolver. `grpcs://` scheme doesn't exist, though, so the default one
is picked. The default resolver (`dns://`) is in turn unable to parse the
5edab9e554/internal/resolver/dns/dns_resolver.go (L405)
The error is `grpcs://192.168.198.248:8081:443: too many colons in address`.
Both problems don't exist in the SDK code, take it from there.
Change-Id: Ia1212050f539162a560796685efdc3f9cfbf80a0
Signed-off-by: Evgenii Stratonikov <e.stratonikov@yadro.com>
They connect to the same endpoints, the only difference is that
connection for synchronization is limited in lifetime and is closed
after the sync is finished. This is probably not intentional, as
synchronization was implemented before cache was introduced.
However, reusing dialTreeService() in sync.go has possible perfomance
implications, so is avoided for now.
Change-Id: I2e37befd783b4d873ff833969f932deded1195be
Signed-off-by: Evgenii Stratonikov <e.stratonikov@yadro.com>
* Move some helpers from `acl/v2` package to `ape`. Also move errors;
* Introduce `Metadata`, `RequestInfo` types;
* Introduce `RequestInfoExtractor` interface and its implementation.
The extractor's purpose is to extract request info based on request
metadata. It also validates session token;
* Refactor ape service - each handler forms request info and pass
necessary fields to checker.
Signed-off-by: Airat Arifullin <a.arifullin@yadro.com>
Tree sync is too much different from GC and rebuild to use the same tag for GC and tree sync.
Change-Id: Ib44d5fa9a88daff507d759d0b0410cc9272e236f
Signed-off-by: Dmitrii Stepanov <d.stepanov@yadro.com>
When we do `object patch` with audit enabled we get several
duplicated entries in logs.
`object patch` request is logged in 2 places:
1. `(*auditPatchStream) CloseAndRecv()` - when the client closes
the request stream or when stream gets aborted.
2. `(*auditPatchStream) Send()` - when stream was NOT aborted.
`Send()` doesn't check if `err != nil` before logging.
It led to to logging on every `Send()` call.
Signed-off-by: Ekaterina Lebedeva <ekaterina.lebedeva@yadro.com>
Added method `ListShardsForObject` to ControlService and to
StorageEngine. It returns information about shards storing
object on the node.
Signed-off-by: Ekaterina Lebedeva <ekaterina.lebedeva@yadro.com>
By default, gRPC fetches TXT report while resolving a domain.
0914bba6c5/internal/resolver/dns/dns_resolver.go (L336)
This leads to a hanging dial if DNS is unavailable, even though the host
may be specified in `/etc/hosts` (hello, localhost!).
SDK client for the main API uses these options by default.
Refs TrueCloudLab/frostfs-sdk-go#342
Signed-off-by: Evgenii Stratonikov <e.stratonikov@yadro.com>
When we do `object put` with audit enabled we get several entries
in logs: with and without object id.
`object put` request is logged in 2 places:
1. `(*auditPutStream) CloseAndRecv()` - when the client closes the
request stream or when stream gets aborted.
2. `(*auditPutStream) Send()` - when stream was NOT aborted.
`Send()` does error check for `ErrAbortStream` because if there
is any other error - CloseAndRecv will not be called and there
won't be any audit log about failed request.
It led to logging on every object chunck put, even if `err == nil`.
Added check for `err != nil` in `Send()` to fix it.
Signed-off-by: Ekaterina Lebedeva <ekaterina.lebedeva@yadro.com>
* A client may open stream to server, not send anything and close
the open stream immediatly. This shouldn't cause a panic;
* Return the error if `s.patcher` is uninitialized. Uninitialized
patcher cannot be closed, this causes a panic.
Signed-off-by: Airat Arifullin <a.arifullin@yadro.com>
If applyOperationStream() exits prematurely, other goroutines will block
on send and errgroup will never finish waiting. In this commit we also
check whether context is cancelled.
Signed-off-by: Evgenii Stratonikov <e.stratonikov@yadro.com>
Constant string `testOwnerID` for these tests has an invalid
format. It has 11 bytes instead of required 25 for `user.ID`.
It worked because:
1. `user.ID` was a byte slice and didn't check length
and format of byte slices decoded from strings.
2. in these tests `testOwnerID` was used only to decode
container owner id and to compare it with owner id encoded
back to string.
Since `user.ID implementation has changed`, the problem arised.
Now `testOwnerID` is valid.
Signed-off-by: Ekaterina Lebedeva <ekaterina.lebedeva@yadro.com>
Current flow is hard to reason about, #1601 is a notorious example of
accidental complexity.
1. Remove multiple nested ifs, use depth=1.
2. Process each status exactly once, hopefully preventing bugs like
#1601.
Signed-off-by: Evgenii Stratonikov <e.stratonikov@yadro.com>
Consider `REP 1 REP 1` placement (selects/filters are omitted).
The placement is `[1, 2], [1, 0]`. We are the 0-th node.
Node 1 is under maintenance, so we do not replicate object
on the node 2. In the second replication group node 1 is under maintenance,
but current caching logic considers it as "replica holder" and removes
local copy. Voilà, we have DL if the object is missing from the node 1.
Signed-off-by: Evgenii Stratonikov <e.stratonikov@yadro.com>
The node can have MAINTENANCE status in the network map, but can also be
ONLINE while responding with MAINTENANCE. These are 2 different code
paths, let's test them separately.
Signed-off-by: Evgenii Stratonikov <e.stratonikov@yadro.com>
