Sreeja created HDDS-15305:
-----------------------------
Summary: ozone admin container list --all returns duplicate
containers due to sort/pagination key mismatch in SCM
Key: HDDS-15305
URL: https://issues.apache.org/jira/browse/HDDS-15305
Project: Apache Ozone
Issue Type: Bug
Reporter: Sreeja
Assignee: Sreeja
The *ozone admin container list --all* command produces duplicate container
entries in its output. The duplicates are non-deterministic across runs, which
containers are duplicated depends on the order in which the SCM returns them,
which can vary.
Root Cause:
The bug is a mismatch between the server-side sort key and the client-side
pagination key Server side (SCMClientProtocolServer.listContainerInternal):The
method filters containers by containerID >= startContainerID, then calls
.sorted() which invokes ContainerInfo.compareTo(). That comparator is defined
as:
{code:java}
private static final Comparator<ContainerInfo> COMPARATOR =
Comparator.comparingLong(info -> info.getLastUsed().toEpochMilli()); {code}
So the batch is returned sorted by lastUsed timestamp, not by containerID.
Client side (ListSubcommand.listAllContainers): The pagination loop advances
the cursor using the last returned element's container ID.
This assumes the last element of the batch has the highest container ID. That
is only true if the batch is sorted by container ID. Since it is actually
sorted by lastUsed, the last element can have a lower container ID than other
elements already in the same batch.
Example: With batch size 40, a batch may return containers with IDs [..., 41,
42, 44, 40] (sorted by lastUsed). The cursor is set to 40 + 1 = 41. The next
batch fetches containers with containerID >= 41, returning [41, 42, 43, 44, 45,
...] — re-fetching 41, 42, and 44.
Fix :
In SCMClientProtocolServer.listContainerInternal, replace the lastUsed-based
sort with a containerID-based sort, aligning the sort key with the pagination
key:
{code:java}
.sorted(Comparator.comparing(ContainerInfo::containerID)){code}
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
