Conversation
The box ships `/etc/netplan/01-netcfg.yaml` declaring an `eth0` that does not exist on this hardware; the NIC is `enp0s3`, configured by `00-installer-config.yaml`. netplan feeds every declared interface into the generated `systemd-networkd-wait-online` drop-in, and that unit is invoked without `--any`, so it waits for `eth0` until its 120 s timeout expires and then fails. `network-online.target`, and therefore `ssh.service`, stays blocked for that whole time. Remove the file during provisioning, with `run: "always"` so existing VMs are corrected on their next `vagrant up` rather than only VMs created after a destroy. Measured on `bento/ubuntu-26.04` 202606.01.0: boot 3:11 before, 1:45 after, with `systemd-networkd-wait-online` down from a 120 s timeout to 178 ms. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Several machines run at once, one `web`/`epub` pair per worktree, and all of them are created from the same box. `vb.linked_clone` makes each machine share a single base disk image instead of taking a full private copy, so the host page cache can serve the blocks they read in common instead of reading each copy separately. Creation-time only: existing machines keep their full copies until they are destroyed and recreated, so this has no effect until then and is not yet exercised. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
VirtualBox creates the storage controller with host I/O caching off, so every guest read goes to the physical disk. With several machines booting near-identical images that is the dominant cost of `vagrant up`: the guest waits tens of seconds for its own virtual disk to appear, and everything else, including `ssh.service`, queues behind it. `LEARN_VM_HOST_IO_CACHE` turns caching on, and `LEARN_VM_STORAGE_CONTROLLER` names the controller to act on, since that name is box-specific. Both are read on every `vagrant up`, so they apply to existing machines. The default is off, matching VirtualBox: with caching on, guest writes may sit in host RAM, so a host crash can corrupt a guest file system. An unrecognized cache value is refused rather than ignored, so a typo cannot silently leave the controller in a state the caller did not ask for. Validation: `useHostIOCache` confirmed to flip in the VM's own configuration; boot time falls by roughly 9x with caching on, and the source-code example build is unchanged either way, as it reads from the synced folders and is otherwise CPU-bound. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
`snapd.apparmor.service` and `snapd.socket` sit on the critical chain to `basic.target`, and so delay `ssh.service` -- the unit Vagrant waits for while it prints "Connection reset. Retrying...". Nothing in these build VMs uses snap. The units are masked rather than removed: masking is idempotent, so the provisioner is safe to re-run, and `systemctl unmask` restores them. `run: "always"` so existing machines are covered too, since they report "Machine already provisioned" and would otherwise never run it. Validation: both machines report `masked` for all four units, a second run is a no-op, and `systemctl is-system-running` still reports `running` with no failed units. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Boot was dominated by two avoidable costs: a stale
eth0stanza in thebase box that blocked startup for a fixed two minutes, and uncached guest
disk reads that made concurrent VMs contend for the host disk. Boot falls
from 3:11 to ~1:45 by default, and to 10-20 s with caching enabled.
/etc/netplan/01-netcfg.yaml. It declares aneth0that does not exist here, and netplan feeds it intosystemd-networkd-wait-online, which is invoked without--anyand soblocks
network-online.target-- andssh.service-- for its full120 s timeout.
LEARN_VM_HOST_IO_CACHE(on/off, defaultoff) andLEARN_VM_STORAGE_CONTROLLERto control host I/O caching on the storagecontroller. Off by default, matching VirtualBox: with caching on, guest
writes may sit in host RAM, so a host crash can corrupt a guest file
system.
a full copy each.
ssh.serviceand are unused in these build VMs.
Only
Vagrantfilechanges. Validated on both machines with four VMsrunning: boot times as above, toolchain verified by compiling and running
a program, and
make test_content_courses_ada-in-practicebyte-identicalagainst an unchanged control VM.
Co-Authored-By: Claude Opus 5 (1M context) noreply@anthropic.com