Memory-mapped files on am EBS volume

@denis2025,

Please find the risk factors when you move the files from SSD to the EBS volume. Let’s unpack the 3 cases:

1. Reading a page from EBS when it’s unavailable

When your program uses a memory-mapped file and tries to read data that isn’t already in memory, the operating system (OS) loads that data from storage.

If the storage is on EBS (Amazon’s network-based disk), this means the OS must fetch the data over the network.

If there’s a temporary network or EBS issue:

The read will pause and your program will hang until the OS gets a response.

If the problem is short-lived, it may recover and continue normally.

If the problem is serious (EBS disconnects, network breaks, etc.), the OS will return an error.

When that happens, your program may get a “bus error” (SIGBUS), which often causes a crash unless it’s handled in code.

In short: reading from an EBS-backed mmap file can freeze or crash your program if the EBS volume goes offline.

2. Flushing dirty pages to an unavailable EBS volume

When the OS needs to save modified data (dirty pages) from memory back to the EBS disk:

It starts a write operation to EBS.

If EBS is slow or has temporary issues, the write will stall while the OS keeps retrying.

If the problem lasts, the OS will return I/O errors.

Depending on the filesystem:

The filesystem may switch to read-only mode to protect data.

Your program may see errors when calling msync() or fsync().

Later writes to the mapped file could again trigger a SIGBUS error.

In short: failed writes can freeze your app, cause crashes, or make the filesystem unstable — though journaling filesystems (like ext4) usually prevent total corruption.

3. Practical implications

Using memory-mapped files on EBS comes with these risks:

Reads or writes can freeze due to network delays or retries.

Your program can crash (SIGBUS) if EBS disconnects.

The filesystem can become read-only or throw errors if EBS stays unavailable.

By comparison, a local SSD is faster and only fails if the hardware breaks — which is rarer and simpler to handle.

Regards,
Nick R
Cloud Team Lead
AccuWeb.Cloud