Devesh Kumar Singh created HDDS-15766:
-----------------------------------------
Summary: Recon - Make Recon OM DB large tar ball transfer reliable
Key: HDDS-15766
URL: https://issues.apache.org/jira/browse/HDDS-15766
Project: Apache Ozone
Issue Type: Task
Components: Ozone Recon
Reporter: Devesh Kumar Singh
Assignee: Devesh Kumar Singh
Recon currently obtains a full OM DB snapshot through a single large HTTP tar
stream. That design works when the transfer completes in one pass, but it
becomes fragile for very large checkpoints or long-lived transfers.
If the response body is interrupted by any of the following:
OM restart / crash / local serving failure,
Recon restart / crash / client-side abort,
proxy, load balancer, firewall, service mesh, or NAT timeout,
transient network reset,
Recon receives an incomplete tar stream and fails with errors such as:
{code:java}
java.io.IOException: Premature EOF
{code}
Existing recon OM DB transfer can be improved in 2 phases:
1. Use similar mechanism what OM follower uses where it downloads multiple tar
balls of bounded size and excludes the downloaded sst files of first tarball in
next tar ball download request.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
