Previously, dhcpcd and bitcoind starting up in parallel could lead to
the following error in bitcoind:
```
bitcoind: libevent: getaddrinfo: address family for nodename not supported
bitcoind: Binding RPC on address 127.0.0.1 port 8332 failed.
bitcoind: Unable to bind any endpoint for
```
After the initial failure, the bitcoind service would always restart successfully.
This race condition, where both applications were simultaneously
manipulating network resources, was only triggered under specific
hardware conditions.
Fix it by running bitcoind after dhcp has started (by running after
`network-online.target`).
This bug and the fix only affect the default NixOS scripted
networking backend.
Currently, public services that publish their onion addresses have a
`requires` dependeny on service `onion-addresses`, so they are restarted
when `onion-addresses` restarts.
This has the downside that most of nix-bitcoin's services are restarted
when a new onion service is added and the system config has the following common settings:
- nix-bitcoin.onionServices.bitcoind.public = true
- nix-bitcoin.operator.enable = true
Sequence of events:
1. onion service is added
2. `onion-addresses` restarts, because the new onion service is made available to `operator`
3. bitcoind restarts
4. all depending services restart
Fix this by using a `wants` dependency.
I've just seen `liquidd` hit the timeout on nixbitcoin.org while
`Loading block index`.
This was probably due to HDD contention while starting services during
boot.
This makes our list definitions mergeable with custom list values
set by users.
Previously, a module error ("value is a string while a list
was expected") was thrown instead.
This commit was partly auto-generated with this script:
#!/usr/bin/env ruby
Dir["**/*.nix"].each do |file|
src = File.read(file)
fixed = src.gsub(/ReadWritePaths *= *(.*?);/) do
"ReadWritePaths = [ #{$1} ];"
end
File.write(file, fixed) if fixed != src
end
This enables generating module option documentation.
This commit was genereated by running the following script inside the
repo root dir:
def add_default_text(file)
src = File.read(file)
src2 = src.gsub(/( = mkOption\s+\{[^{]*?)(\n\s+default = )(.*?);$(.*?\})/m) do |str|
pre, defaultVar, default, post = Regexp.last_match.captures
replacement =
if !post.include?('defaultText =')
if default =~ /\bpkgs\b/
defaultText = default.lines.length == 1 ? default : "(See source)"
"#{pre}#{defaultVar}#{default};#{defaultVar.sub('default', 'defaultText')}#{defaultText.inspect};#{post}"
end
end
replacement or str
end
File.write(file, src2) if src2 != src
end
Dir["modules/**/*.nix"].each do |f|
next if File.basename(f) == "nix-bitcoin.nix"
add_default_text f
end
Split `enforceTor` into `tor.proxy` and `tor.enforce`.
By enabling `tor.proxy` without `tor.enforce`, a service can accept
incoming clearnet connections.
E.g., this allows setting up a Tor-proxied bitcoind node that accepts
RPC connections from LAN.
This allows whitelisting local services without implicitly
whitelisting all inbound onion connections, which would happen when
setting bitcoind/liquidd option `whitelist=localhost`.
Used by electrs and nbxplorer, which requires the unsafe `mempool`
permission.
This removes the module-level dependency from onion-services to
bitcoind.
Due to the `or false` fallback, there's no dependency added in
the reverse direction.
In particular, this allows us to not add a dependency on liquidd in
the following commit.
This re-enables onion tagging while still supporting untagged connections.
Onion sockets are not yet supported in the latest liquidd/elements
version 0.18.1.12 available on nixpkgs.
- bitcoind: Remove obsolete defaultText
- clightning: Fix description
Option `address` can't be used to specify a socket path because it's
used explicitly as an IP address in many places.
- lnd: Break up overlong line
This is required by commit `services: support 0.0.0.0/:: in `address` options`
- nix-bitcoin.nix: Formatting
- secrets: Improve descriptions
These are insignificant, generic options; place them above readonly options.
We already do this in other services.
Also move user/group config to bottom in spark-wallet.
This greatly improves readability and makes it easier to discover options.
This commit was genereated by running the following script inside the
repo root dir:
#!/usr/bin/env ruby
def transform(src)
return false if src.include?('inherit options;')
success = false
options = nil
src.sub!(/^ options.*?^ }.*?;/m) do |match|
options = match
" inherit options;"
end
return false if !options
src.sub!(/^with lib;\s*let\n+/m) do |match|
success = true
<<~EOF
with lib;
let
#{options}
EOF
end
success
end
Dir['modules/**/*.nix'].each do |f|
src = File.read(f)
if transform(src)
puts "Changed file #{f}"
File.write(f, src)
end
end
`generate-secrets` is no longer a monolithic script. Instead, it's
composed of the values of option `nix-bitcoin.generateSecretsCmds`.
This has the following advantages:
- generate-secrets is now extensible by users
- Only secrets of enabled services are generated
- RPC IPs in the `lnd` and `loop` certs are no longer hardcoded.
Secrets are no longer automatically generated when entering nix-shell.
Instead, they are generated before deployment (via `krops-deploy`)
because secrets generation is now dependant on the node configuration.
Due to a possible NixOS bug, this commit has no effect on NixOS 20.09
where `RestrictAddressFamilies` is a no-op.
It's only relevant for NixOS unstable with cgroups v2.
bitcoind+zmq: instead of allowing all address families, only add the required
AF_NETLINK family.
lnd: lnd only runs a zmq client, not a server, therefore it requires
no additional address families.
lightning-pool, clightning-plugin-zmq: add AF_NETLINK.
- Improve readability by using minutes
- set `TimeoutStopSec` like in bitcoin/contrib/init/bitcoind.service.
Stopping bitcoind can exceed the default timeout during IBD.
With krops or nixops the secrets target is always restarted during
the deployment process.
This previously caused unnecessary restarts of all nix-bitcoin services.
Systemd's `Description` option is a misnomer (as confessed by `man systemd.unit`):
Its value is used by user-facing tools in place of the unit file name, so this option
could have been more aptly named `label` or `name`.
`Description` should only be set if the unit file name is not sufficient for naming a unit.
This is not the case for our services, except for `systemd.services.nb-netns-bridge`
whose description has been kept.
As an example how this affects users, weird journal lines like
```
nb-test systemd[1]: Starting Run clightningd...
```
are now replaced by
```
nb-test systemd[1]: Starting clightning.service...
```
