better error handling for run hooks

[drewbanin]

query errors in on-run-start and on-run-end hooks aren't appropriately caught/handled.

Example backtrace:

Traceback (most recent call last):
  File "/usr/local/bin/dbt", line 11, in <module>
    load_entry_point('dbt==0.8.2', 'console_scripts', 'dbt')()
  File "/usr/local/Cellar/dbt/0.8.2/libexec/lib/python3.6/site-packages/dbt/main.py", line 37, in main
    handle(args)
  File "/usr/local/Cellar/dbt/0.8.2/libexec/lib/python3.6/site-packages/dbt/main.py", line 63, in handle
    res = run_from_args(parsed)
  File "/usr/local/Cellar/dbt/0.8.2/libexec/lib/python3.6/site-packages/dbt/main.py", line 117, in run_from_args
    result = task.run()
  File "/usr/local/Cellar/dbt/0.8.2/libexec/lib/python3.6/site-packages/dbt/task/run.py", line 18, in run
    results = runner.run_models(self.args.models, self.args.exclude)
  File "/usr/local/Cellar/dbt/0.8.2/libexec/lib/python3.6/site-packages/dbt/runner.py", line 800, in run_models
    should_run_hooks=True)
  File "/usr/local/Cellar/dbt/0.8.2/libexec/lib/python3.6/site-packages/dbt/runner.py", line 770, in run_types_from_graph
    should_execute)
  File "/usr/local/Cellar/dbt/0.8.2/libexec/lib/python3.6/site-packages/dbt/runner.py", line 627, in execute_nodes
    self.run_hooks(profile, flat_graph, dbt.utils.RunHookType.End)
  File "/usr/local/Cellar/dbt/0.8.2/libexec/lib/python3.6/site-packages/dbt/runner.py", line 496, in run_hooks
    adapter.execute_all(profile=profile, sqls=compiled_hooks)
  File "/usr/local/Cellar/dbt/0.8.2/libexec/lib/python3.6/site-packages/dbt/adapters/default.py", line 547, in execute_all
    connection, _ = cls.add_query(profile, sql, model_name)
  File "/usr/local/Cellar/dbt/0.8.2/libexec/lib/python3.6/site-packages/dbt/adapters/default.py", line 526, in add_query
    cursor.execute(sql)
psycopg2.ProgrammingError: relation "dbt_jthandy.some_table" does not exist

[drewbanin]

This particular error shouldn't happen anymore. If there's an error in an on-run-start or on-run-end hook, dbt will hard exit, but will not show a backtrace.

$ dbt run
Encountered an error:
schema "bad_schema" does not exist
LINE 1: create table if not exists bad_schema.hook_test (id int)
                                   ^
$

@jthandy what do you think should happen if an on-run-start hook fails? I don't think we can skip dependents as we do with models, since every model is a dependent! My instinct is that cancelling the run makes sense, but the above output would benefit from more info for the end-user.

[jthandy]

I don't think error messages (even ones that are passed through from the database) are a bad thing here. The only thing that would be helpful is if dbt could tell the user what it was attempting to run when it encountered that error message. Could it say:

$ dbt run
Encountered an error **while running pre-hook**:
schema "bad_schema" does not exist
LINE 1: create table if not exists bad_schema.hook_test (id int)
                                   ^