lightningd: Added --alt-subdaemon command to allow alternate subdaemons.

[ksedgwic]

This PR adds --alt-subdaemon to facilitate substituting a standard subdaemon with an alternate.

Examples:

# Use remote_hsmd instead of lightning_hsmd for signing:
lightningd --alt-subdaemon=lightning_hsmd:remote_hsmd ...

# Same as previous but daemon is located w/ absolute path:
lightningd --alt-subdaemon=lightning_hsmd:/some/where/remote_hsmd ...

[darosior]

That's a cool idea. Just some (mostly style) nits, I'll leave the conceptual review to the others.

[devrandom]

seems like there was an intermittent failure in the build for 6c727d6

[arowser]

Nice pr, @rustyrussell @cdecker another idea: if change the message channels between sub daemons from stdio to socket , then the subdaemon could run on different machines, even run remote-hsmd on a real remote HSM.

[ZmnSCPxj]

@arowser how would we start the processes remotely? I suggest rather that such setups can be done with a bridge daemon that replaces --alt-subdaemon locally, then just splice (on Linux; otherwise just copy-in-userspace) from the stdio to a local pipe and to a remote socket (on Linux, pipes are "just" representations of kernel buffers, so the local pipe is only needed to properly adapt to any combination of endpoint devices), with the bridge daemon knowing how to properly initiate starting the remote process.

[rustyrussell]

Some trivial changes might be nice, but this is fine as is, too!

Thanks!

[rustyrussell]

@ksedgwic Thanks for your patience!

Almost there: let's not skip the final d (my typo, sorry!):

diff --git a/doc/lightningd-config.5.md b/doc/lightningd-config.5.md
index c72121bff..9d6548505 100644
--- a/doc/lightningd-config.5.md
+++ b/doc/lightningd-config.5.md
@@ -109,12 +109,12 @@ by *--conf*.
 
  **subdaemon**=*SUBDAEMON*:*PATH*
 Specifies an alternate subdaemon binary.
-Current subdaemons are *channel*, *closing*,
-*connect*, *gossip*, *hsm*, *onchain*, and *opening*.
+Current subdaemons are *channeld*, *closingd,
+*connectd*, *gossipd*, *hsmd*, *onchaind*, and *openingd*.
 If the supplied path is relative the subdaemon binary is found in the
 working directory. This option may be specified multiple times.
 
- So, **subdaemon=lightning_hsmd:remote_signer** would use a
+ So, **subdaemon=hsmd:remote_signer** would use a
 hypothetical remote signing proxy instead of the standard *lightning_hsmd*
 binary.
 
diff --git a/lightningd/lightningd.c b/lightningd/lightningd.c
index 2f658affc..a43f3f07a 100644
--- a/lightningd/lightningd.c
+++ b/lightningd/lightningd.c
@@ -316,11 +316,9 @@ static void memleak_help_alt_subdaemons(struct htable *memtable,
 
 const char *subdaemon_path(const tal_t *ctx, const struct lightningd *ld, const char *name)
 {
-	/* Strip the leading "lightning_" and following 'd' before looking
-	 * in alt_subdaemons
-	 */
-	assert(strlen(name) > 10 + 1);	/* len("lightning_") + len("d") */
-	const char *short_name = tal_strndup(ctx, name + 10, strlen(name) - 11);
+	/* Strip the leading "lightning_" before looking in alt_subdaemons */
+	assert(strlen(name) > strlen("lightning_"));
+	const char *short_name = tal_strdup(ctx, name + strlen("lightning_"));
 
 	/* Is there an alternate path for this subdaemon? */
 	const char *dpath;

And if (strncmp(sdname, subdaemons[i] + 10, strlen(sdname)) == 0)
should probably be strcmp, e.g.: if (streq(sdname, subdaemons[i] + strlen("lightning_"))
otherwise you can do --subdaemon=c:mydaemon and replace all subdaemons starting with c, which is not what we want?

Cheers!

[ksedgwic]

I think that was a spurious CI failure (timeout), force pushed to try again.

[cdecker]

ACK 95b5bdb