agd fails with "Cannot find dependency ..." in systemd due to lack of file descriptors

[dckc]

Describe the bug

When starting agd in systemd, it fails with Error#1: config.bundles.coreProposal2_5: Cannot find dependency picomatch ...

To Reproduce

NodesGuru reports:

# Build software
CI_GIT_NAME=Agoric
CI_GIT_FOLDER=agoric-sdk
CI_BIN_VER=ea8c1c64911b4c58fb43635b25e17e3d50d0cf2a
CI_BIN_NAME=agd

cd $HOME
git clone https://github.com/${CI_GIT_NAME}/${CI_GIT_FOLDER}.git
cd $HOME/${CI_GIT_FOLDER}
git fetch --all
git checkout ${CI_BIN_VER}

git submodule update
find . -name node_modules |xargs rm -rf
sudo apt update
curl https://deb.nodesource.com/setup_16.x | sudo bash
sudo apt install -y nodejs gcc g++ make < "/dev/null"
curl -sL https://dl.yarnpkg.com/debian/pubkey.gpg | gpg --dearmor | sudo tee /usr/share/keyrings/yarnkey.gpg >/dev/null
echo "deb [signed-by=/usr/share/keyrings/yarnkey.gpg] https://dl.yarnpkg.com/debian stable main" | sudo tee /etc/apt/sources.list.d/yarn.list
sudo apt update && sudo apt install yarn
yarn install --force
yarn build
(cd $HOME/${CI_GIT_FOLDER}/packages/cosmic-swingset && make)

sudo cp $HOME/go/bin/${CI_BIN_NAME} /usr/local/bin/${CI_BIN_NAME}

${CI_BIN_NAME} version

Then in a systemd unit:

$ cat /etc/systemd/system/agd.service
[Unit]
Description=Agoric Node
After=network-online.target
[Service]
User=ubuntu
ExecStart=/home/ubuntu/agoric-sdk/bin/agd start --address tcp://0.0.0.0:55658 --grpc-web.address 0.0.0.0:12091 --grpc.address 0.0.0.0:12090 --p2p.laddr tcp://0.0.0.0:53956 --rpc.laddr tcp://127.0.0.1:56657 --home /home/ubuntu/.agoric
Environment=PATH="/home/ubuntu/go/bin:/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin:/usr/games:/usr/local/games:/snap/bin:/usr/local/go/bin:/home/ubuntu/go/bin:/usr/local/go/bin:/home/ubuntu/go/bin:/usr/local/go/bin:/home/ubuntu/go/bin:/usr/local/go/bin:/home/ubuntu/go/bin:/usr/local/go/bin:/home/ubuntu/go/bin:/usr/local/go/bin:/home/ubuntu/go/bin:/usr/local/go/bin:/home/ubuntu/go/bin:/usr/local/go/bin:/home/ubuntu/go/bin:/usr/local/go/bin:/home/ubuntu/go/bin"
Restart=always
RestartSec=3
LimitNOFILE=4096
[Install]
WantedBy=multi-user.target

Expected behavior

agd works as a systemd service

Platform Environment

• what OS are you using? what version of Node.js?
• is there anything special/unusual about your platform?
• what version of the Agoric-SDK are you using? (run git describe --tags --always)

ea8c1c6

Additional context

agoricdev-18

note discord #devnet thread

Screenshots

stack trace from 0xAN | Nodes.Guru:

portHandler threw (Error#1)
Error#1: config.bundles.coreProposal2_5: Cannot find dependency picomatch for file:///home/ubuntu/agoric-sdk/node_modules/ava/
  at packages/SwingSet/src/controller/initializeSwingset.js:538:15
  at async Promise.all (index 18)
  at async processGroup (packages/SwingSet/src/controller/initializeSwingset.js:541:27)
  at async initializeSwingset (packages/SwingSet/src/controller/initializeSwingset.js:570:43)
  at async ensureSwingsetInitialized (packages/cosmic-swingset/src/launch-chain.js:160:5)
  at async buildSwingset (packages/cosmic-swingset/src/launch-chain.js:165:3)
  at async launch (packages/cosmic-swingset/src/launch-chain.js:307:52)
  at async launchAndInitializeSwingSet (packages/cosmic-swingset/src/chain-main.js:453:15)
  at async toSwingSet (packages/cosmic-swingset/src/chain-main.js:670:20)
Cannot initialize Controller Error: config.bundles.coreProposal2_5: Cannot find dependency picomatch for file:///home/ubuntu/agoric-sdk/node_modules/ava/
agd.service: Main process exited, code=exited, status=1/FAILURE

logs from Syd | FR Staking Community:

May 20 1359 ubuntu-8gb-hel1-1 systemd[1]: Started Agoric Cosmos daemon.
May 20 1300 ubuntu-8gb-hel1-1 agd[2190252]: 2023/05/20 1300 Running SwingSet until bootstrap is ready
May 20 1300 ubuntu-8gb-hel1-1 agd[2190252]: Loading slog sender modules: @Agoric/telemetry/src/flight-recorder.js
May 20 1300 ubuntu-8gb-hel1-1 agd[2190252]: 2023-05-20T1300.491Z launch-chain: Launching SwingSet kernel
May 20 1317 ubuntu-8gb-hel1-1 agd[2190252]: portHandler threw (Error#1)
May 20 1317 ubuntu-8gb-hel1-1 agd[2190252]: Error#1: config.bundles.coreProposal2_5: Cannot find dependency picomatch for file:///home/ubuntu/agoric-sdk/node_modules/ava/
May 20 1317 ubuntu-8gb-hel1-1 agd[2190252]:   at packages/SwingSet/src/controller/initializeSwingset.js:538:15
May 20 1317 ubuntu-8gb-hel1-1 agd[2190252]:   at async Promise.all (index 18)
May 20 1317 ubuntu-8gb-hel1-1 agd[2190252]:   at async processGroup (packages/SwingSet/src/controller/initializeSwingset.js:541:27)
May 20 1317 ubuntu-8gb-hel1-1 agd[2190252]:   at async initializeSwingset (packages/SwingSet/src/controller/initializeSwingset.js:570:43)
May 20 1317 ubuntu-8gb-hel1-1 agd[2190252]:   at async ensureSwingsetInitialized (packages/cosmic-swingset/src/launch-chain.js:160:5)
May 20 1317 ubuntu-8gb-hel1-1 agd[2190252]:   at async buildSwingset (packages/cosmic-swingset/src/launch-chain.js:165:3)
May 20 1317 ubuntu-8gb-hel1-1 agd[2190252]:   at async launch (packages/cosmic-swingset/src/launch-chain.js:307:52)
May 20 1317 ubuntu-8gb-hel1-1 agd[2190252]:   at async launchAndInitializeSwingSet (packages/cosmic-swingset/src/chain-main.js:453:15)
May 20 1317 ubuntu-8gb-hel1-1 agd[2190252]:   at async toSwingSet (packages/cosmic-swingset/src/chain-main.js:670:20)
May 20 1317 ubuntu-8gb-hel1-1 agd[2190252]: Cannot initialize Controller Error: config.bundles.coreProposal2_5: Cannot find dependency picomatch for file:///home/ubuntu/agoric-sdk/node_modules/ava/
May 20 1317 ubuntu-8gb-hel1-1 systemd[1]: agd.service: Main process exited, code=exited, status=1/FAILURE
May 20 1317 ubuntu-8gb-hel1-1 systemd[1]: agd.service: Failed with result 'exit-code'.
May 20 1320 ubuntu-8gb-hel1-1 systemd[1]: agd.service: Scheduled restart job, restart counter is at 34452.
May 20 1320 ubuntu-8gb-hel1-1 systemd[1]: Stopped Agoric Cosmos daemon.

[dckc]

reported work-around: run agd outside systemd

[dckc]

related:

• separate agd build process from runtime files #6455

[dckc]

diagnosis: bundling ran into file descriptor limit

At start-up, agd does a lot of bundling of JavaScript modules. Outside of systemd, ubuntu has a limit around 1 million. In the reported configuration, we see:

LimitNOFILE=4096

We should reduce the required number of simultaneous file descriptors in due course, but in the mean time

work-around: increase file descriptor limit

Using 64K file descriptors seems to relieve the symptoms:

LimitNOFILE=65536

[warner]

endojs/endo#1593 is the long-term fix for this, to either limit the bundle-source parallelism, and/or react to EMFILE by deferring the open() until some other FD has been closed.

[kriskowal]

I’ve landed a fix for endojs/endo#1593 in Endo and this should be a thing of the past when we next sync Endo releases with Agoric SDK.

[dckc]

@kriskowal is this now a thing of the past? It's in the upgrade-11 release notes.

[kriskowal]

No, I have not yet successfully synced Endo with Agoric SDK. This is more likely to land in upgrade-12.

[kriskowal]

I believe this is now a thing of the past. Please reöpen if symptoms persist.