db_set_intvar:disk I/O error when running as daemon on macOS

[Sjors]

Using 3ad078d (from #1090) to run a fresh testnet node on macOS as a daemon:

2018-02-24T14:54:29.419Z lightningd(6078): Adding block 1285443: 00000000000007a0293aba4a64af93ba7e50bf44a652d9afa8245c7df7b9a0aa
2018-02-24T14:54:29.427Z lightningd(6078): db_set_intvar:disk I/O error:UPDATE vars SET val='1285443' WHERE name='last_processed_block';:disk I/O error
2018-02-24T14:54:29.427Z lightningd(6078): FATAL SIGNAL 6 RECEIVED
2018-02-24T14:54:29.427Z lightningd(6078): error getting backtrace: lightningd/lightningd (2)
2018-02-24T14:54:29.427Z lightningd(6078): error getting backtrace: failed to read executable information (-1)

Wiping the .lightning dir and restarting yields the same result. Not running as daemon seems to prevent it.

Possibly related to #1005.

I uploaded the full log and slite3 database:
node.zip

[rustyrussell]

@Sjors Ah, good debugging! Indeed, keeping the db connection open across the fork is not recommended, according to the sqlite3 FAQ.

We deliberately don't daemonize() early, because it's painful if there are startup issues: you really want a failure to stderr in that case. The cost is that some things are delicate and get upset; this is one of them.