Sergey Soldatov created HDDS-15567:
--------------------------------------
Summary: Streaming block PT10S timeout on backward/large-forward
seeks
Key: HDDS-15567
URL: https://issues.apache.org/jira/browse/HDDS-15567
Project: Apache Ozone
Issue Type: Bug
Components: Ozone Client
Affects Versions: 2.1.0
Reporter: Sergey Soldatov
Assignee: Sergey Soldatov
It's a follow-up for HDDS-15422
When a client reads a block via the streaming read path
(StreamBlockInputStream) and the consumer seeks backward — e.g. the HBase HFile
scanner repositioning after a bloom filter check — the DataNode keeps streaming
responses for the previous (now-abandoned) offset range. The client has moved
to the new offset and ignores those in-flight responses, so the HTTP/2 receive
window fills up and is never drained.
With the receive window full, the server blocks on send. When the client
subsequently needs the response for the new offset, the server can't deliver
it, the read never completes, and the operation fails with a PT10S timeout. The
same stall occurs on large forward seeks, where data streamed for the skipped
range fills the window and blocks the response at the new position.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
