github awslabs/mountpoint-s3 mountpoint-s3-1.25.0
mountpoint-s3 v1.25.0
on GitHub

38 minutes ago

Breaking changes

  • Amazon Linux 2 reached end of support on June 30, 2026 and is no longer tested. Mountpoint for Amazon S3 releases may continue to install and run on it, but we recommend upgrading to Amazon Linux 2023. (#2033)
  • Mountpoint now fails to start when AWS_MAX_ATTEMPTS is not a positive integer. Previously 0 selected 4 attempts and other invalid values were silently ignored. (#1996)

Other changes

  • Fix Mountpoint failing to start with a 30 second timeout when --cache contains a large cache left over from a previous run. Mountpoint no longer enforces a timeout while it removes the old cache, so starting can take as long as the removal does, and it prints a warning if the removal takes longer than 30 seconds. The rest of the mount still fails if it makes no progress for 30 seconds. (#2024)
  • Detect the EC2 instance type from the instance's DMI tables before querying IMDS. Where IMDS is disabled or unreachable, the default target throughput now matches the instance's network bandwidth instead of falling back to 10 Gbps. This can increase memory and network usage; use --maximum-throughput-gbps 10 to keep the previous target. (#2006)
  • Add the unstable UNSTABLE_MOUNTPOINT_DISABLE_RETRY_QUOTA environment variable. When set to true, Mountpoint keeps retrying throttled requests up to the maximum number of attempts even while S3 keeps throttling, instead of failing them without retrying once its retry quota runs out. See the configuration documentation. (#2032)
  • Fix AWS_MAX_ATTEMPTS=1 still retrying failed S3 requests up to 3 times. (#1996)
  • Fix Mountpoint failing to start when AWS_MAX_ATTEMPTS is set to more than 64. (#2010)
  • Use a single event loop thread to query the EC2 instance metadata service (IMDS) at startup, instead of a number of threads that scales with the CPU count. (#2034)
  • Fix memory allocator background threads not running when Mountpoint runs in background mode (the default, without --foreground). These threads return freed memory to the operating system, which otherwise happens on the threads serving file system requests. (#1999)
  • Fix S3 CRT client metrics (s3.client.*) not updating on every metrics publication cycle for requests lasting longer than five seconds. These gauges are now sampled when metrics are published, rather than only when a new S3 request starts. (#1957 by @ziwon)
  • Fix files and directories created through Mountpoint being reported as missing after an earlier lookup of the same name failed, when negative metadata caching is enabled (--metadata-ttl or --negative-metadata-ttl). Lookups could return ENOENT until the negative cache entry expired once the kernel had forgotten the new inode. (#1985 by @samridh10exe)
  • Fix existing directories incorrectly appearing as missing when S3 returns an empty page of listing results. (#1954 by @cayoub-oai)
  • Fix directory listings ending early when S3 returns an empty page of listing results. readdir now keeps listing until it finds entries or reaches the last page. (#2023 by @Manishram-ai)
  • Fix a deadlock that could make the file system hang when the kernel forgets a file or directory while another file in the same directory is renamed, or while the directory is opened for listing with files still being written in it. (#1995)
  • Make --incremental-upload consistently check for concurrent writes when appending (i.e. without O_TRUNC), even at offset 0. Previously, appending to an existing empty object could silently overwrite another writer's changes instead of failing with EIO. (#1970 by @yerzhan7)
  • Update the internal S3 client to use the latest release of the AWS Common Runtime (CRT) libraries. (#1967, #2010, #2020)

Binaries

x86_64

arm64

Check out latest releases or
releases around awslabs/mountpoint-s3 mountpoint-s3-1.25.0

Don't miss a new mountpoint-s3 release

NewReleases is sending notifications on new releases.

Get notifications