[DOCS-11753] macOS 10.13 (APFS) may not be crash safe Created: 30/May/18  Updated: 30/Oct/23  Resolved: 15/Jun/18

Status: Closed
Project: Documentation
Component/s: Server
Affects Version/s: None
Fix Version/s: Server_Docs_20231030

Type: Task Priority: Major - P3
Reporter: Michael Cahill (Inactive) Assignee: Kay Kim (Inactive)
Resolution: Fixed Votes: 0
Labels: installation, macos-high-sierra
Remaining Estimate: Not Specified
Time Spent: Not Specified
Original Estimate: Not Specified

Issue Links:
Related
is related to WT-3724 Log an error if flushing with F_FULLS... Closed
is related to WT-4018 Checkpoint integrity test indicates M... Closed
is related to WT-4112 test failure in recovery test - rando... Closed
is related to SERVER-33002 Platform Support: add MacOS 10.13 (Hi... Closed
Participants:
Days since reply: 5 years, 36 weeks, 6 days ago
Epic Link: DOCS: 4.0 Server

 Description   

Description

We do not yet have stable testing on macOS 10.13, which uses Apple's new APFS filesystem.

In particular, tests of WiredTiger's crash safety fail when run against the macOS 10.13 test system available in Evergreen.

Community users and some MongoDB engineers have been running MongoDB on macOS 10.13 for months without seeing major issues, but the test failures suggest that data may be lost if mongod is not shutdown cleanly, or if the system is powered down while mongod is busy.

Scope of changes (files that need work and how much)

Add an admonition (Warning?) that mongod processes with datafiles on an APFS (10.13+) filesystem may lose data during unclean shutdowns (e.g. power failure, or system shutdown while mongod is busy). 

Possibly point to WT-4018 for discussions and remediation. 

May need a second linked DOCS ticket to update this admonition if a fix is released

Impact to other docs outside of this product

We might consider having a warning on the Cloud Manager and Ops Manager OSX Agent Installation pages w.r.t. 10.13+. The Troubleshooting pages may be another option.

MVP (work and date?)

At minimum, admonition on OSX installation pages. Depending on whether 4.0 includes a fix, may want to then proceed to update the FAQ and production notes.

Coordinate with cloud team on where best to update OpsManager/CloudManager docs.

Resources (e.g. Scope Docs, Invision)

See Linked Tickets 



 Comments   
Comment by Michael Cahill (Inactive) [ 01/Jun/18 ]

ravind.kumar

1) This would affect all versions of MongoDB available for 10.13+, correct?

Yes.

2) From WT-4018, We would like to have a fix ready for 4.0.0, but are dependent on Apple providing more guidance on durably flushing data in APFS. Would we backport this fix, or would this be a 4.0+ only fix (at least in the short term)

We would backport the fix to whatever versions of MongoDB are supported on APFS.

3) Would running the mongod were the dbPath is a non-APFS filesystem resolve this issue?

Maybe. This is one of the questions I'd like to put to an Apple engineer.

4) Are there any advantages to the APFS filesystem that users may want to weigh against the inconsistent crash safety behavior?

According to https://support.apple.com/en-us/HT208033, "When you install macOS High Sierra on the Mac volume of a solid-state drive (SSD) or other all-flash storage device, that volume is automatically converted to APFS."

So if we wanted to give any advice, it would probably be more along the lines of "stick with 10.12 if you need crash safety from MongoDB".

Also note that some searching suggests that later updates to 10.13 (specifically 10.13.4) fix some APFS bugs (e.g., https://github.com/docker/for-mac/issues/2625). We don't yet have solid information about whether upgrading resolves the issues we have seen in testing.

5) It seems like we should generally recommend against 10.13+ for anything outside of development/evaluation environments.

That would be my advice at the moment until WT-4018 is resolved.

Comment by Michael Cahill (Inactive) [ 30/May/18 ]

cc ramon.fernandez, alexander.gorrod, ian.whalen

Generated at Thu Feb 08 08:03:34 UTC 2024 using Jira 9.7.1#970001-sha1:2222b88b221c4928ef0de3161136cc90c8356a66.