Known Issues

These are the current known issues for the Python SDK.

Potential memory leak issue with repeatedly creating new signers/clients with Instance Principals, Resource Principal, and Delegation token auth. Cloud Shell is known to be affected

Details: When repeatedly creating new signers/client objects with Instance Principals, Resource Principal, and Delegation token auth, some underlying objects are not cleared from memory and causes a memory leak. From our tests, the rate of memory growth is ~10 MiB/hour when only client/signer objects are created in an infinite loop. Cloud Shell is affected by the same issue when new client objects are created repeatedly using the Python SDK. This seems to be coming from an underlying memory leak in the requests package.

Workarounds: These workarounds are known to mitigate the issue:

  1. Avoid creating new signer/client objects and reuse them if possible. If you are using delegation token auth, and need to update the value of the delegation token you can do it this way to update an existing signer, instead of creating a new signer:
from oci.auth.signers import InstancePrincipalsDelegationTokenSigner
from oci.object_storage import ObjectStorageClient

# Create signer and client
delegation_token_signer = InstancePrincipalsDelegationTokenSigner(delegation_token="delegation_token_value")
client = ObjectStorageClient(config={}, signer=delegation_token_signer)

# Update the delegation token on the client
client.base_client.signer.delegation_token="new_delegation_token_value"
  1. Use the raw request signer.

Potential data corruption with Python SDK on UploadManager.upload_stream() in FIPS mode and Cloud Shell (versions 2.53.0 and below)

Details: When using the Python SDK to perform stream upload operations you may encounter an issue with data corruption if you are using UploadManager.upload_stream. This issue is known to affect the customers using FIPS mode and Cloud Shell.

Impacted Versions: v2.53.0 and below

Direct link to this issue: Potential data corruption issue for multipart stream upload for OCI Python SDK

Performance regression in oci v2.38.4 and above

In OCI Python SDK version 2.38.4 and later versions, you may encounter performance regressions when using the SDK with some OCI services. The regression surfaces itself as a 3-second increase in latency in SDK operations made to some OCI services. This issue has been confirmed to impact the OCI Streaming service, and likely impacts the following services:

  • Email Delivery
  • Health Checks
  • NoSQL Database Cloud
  • Registry
  • Generic Artifacts
  • Web Application Acceleration
  • Security service

This list is not comprehensive – it is possible you may encounter the issue against other OCI services as well. The issue has been confirmed to NOT affect the OCI Object Storage and Functions services.

We are actively working on fixing this issue, but in the mean time, if you are experiencing this issue there are some workarounds you can do:

  • Recommended: Disable support for the OCI Python SDK’s Expect: 100-Continue feature by adding the following environment variable OCI_PYSDK_USING_EXPECT_HEADER to FALSE. For example, in Bash, export OCI_PYSDK_USING_EXPECT_HEADER=FALSE
  • Or, use OCI Python SDK version 2.38.3 or earlier
  • Or, use the OCI Python SDK’s raw request signer to make requests to OCI services for any requests where you are experiencing performance degradation. An example on how to use the raw request signer is here.

If you have any questions, please feel free to comment on this GitHub issue and we will be happy to help.

create_job_and_wait_for_state() fails with KeyError for ResourceManagerClientCompositeOperations (versions 2.20.0 and above)

ResourceManagerClientCompositeOperations.create_job_and_wait_for_state() fails with KeyError: ‘opc-work-request-id’.

Workaround: Use create_stack() and then implement waiters. You can find an example to use waiters here.

UploadManager.upload_stream() raises MultipartUploadError in oci v2.23.2

UploadManager.upload_stream() raises MultipartUploadError when a timeout is set on the underlying object storage client, and the operation takes more than the read timeout to complete. Prior to v2.23.2, we were overwriting the timeout to None in the operations (please see this known issue). The default timeout is a read timeout of 60 seconds, hence this scenario will be triggered by default in v2.23.2 on any use of this operation where the operation takes 60 or more seconds to complete. You can work around the issue by explicitly setting the timeout to None. For example,

client.base_client.timeout = None

This issue has been fixed in oci v2.23.3

UploadManager generates ssl3_write_pending error when a read timeout is set for the Object Storage client

Update: This issue has partially been fixed in v2.23.2. This issue still may exist with using Python versions < 2.7.9. If you do encounter this issue, please consult the workaround mentioned below.

Update: With v2.18.0 we handle the object storage client with default timeout values (connect timeout = 10 secs and read timeout = 60 secs), by overwriting the timeout to None in the operations.

PLEASE NOTE that the operations are NOT thread-safe, and you should provide the UploadManager class with its own Object Storage client that isn’t used elsewhere.

Details: UploadManager generates the following error when a read timeout is set for the Object Storage client.

OpenSSL.SSL.Error: [('SSL routines', 'ssl3_write_pending', 'bad write retry')]

Workaround: Do not set the read timeout for the Object Storage client. There are a two ways to do so:

  • Create an Object Storage client without setting a timeout.
  • Clear the timeout on an already initialized Object Storage client by setting the timeout of the base_client to None.
client.base_client.timeout = None

If you need to retain the connection timeout while clearing the read timeout, you can do so by setting the timeout to a tuple. The first item in the tuple is the connection timeout and the second is the read timeout. In the following example, the connection timeout is set to 90.0 seconds and the read timeout is set to infinite, or no timeout.

client.base_client.timeout = (90.0, None)

Potential data corruption with Python SDK on binary upload (versions 2.8.0 and below)

Details: When using the Python SDK to perform binary upload operations you may encounter an issue with data corruption if retries are enabled or if you are using UploadManager.upload_file.

Impacted Versions: v2.8.0 and below

Direct link to this issue: Potential data corruption with Python SDK on binary upload

Default timeout not getting set in the clients (versions 2.17.2 and below)

The default timeout values (connect timeout = 10 secs and read timeout = 60 secs) we not getting set in the clients and remained None (infinite timeout). This has been fixed in v2.18.0.

Some BlockStorage composite operations throw a 404/NotAuthorizedOrNotFound for Cross Region operations

Details: The copy_boot_volume_backup_and_wait_for_state() and copy_volume_backup_and_wait_for_state() from the BlockStorage Client Composite operations throw a 404/NotAuthorizedOrNotFound when copying a backup from one region to another even though the operation succeeds eventually.

Impacted Versions: All

Workaround: Instead of using the composite operations, use two different clients for this operation; one client in the Source Region to send the request for copying the backup from Region A to Region B, and a second client in Destination region to wait for the backup to become available. See this for an example.

Direct link to this issue: Some BlockStorage composite operations throw a 404/NotAuthorizedOrNotFound for Cross Region operations