Flow tests don't catch v02 post format regressions

[robjarawan]

what

PR MetPX/sarracenia#1604 (shallow copy in putNewMessage) broke v02 posting with an AttributeError, but the dynamic_flow tests didn't catch it. Reid found it on a live system.

The sender config tsource2send_f50.conf does set post_topic_prefix v02.post, so in theory it should exercise the v02 export path. But this regression shipped without any test failure.

the gap

The flow_check.sh validation counts messages and compares totals, but doesn't verify that posted messages are actually well-formed v02. If the sender crashes on the v02 export, the messages just don't get posted, and the test might still pass if the counts happen to match from other paths.

There's no test that:

1. Posts a message explicitly in v02 format
2. Subscribes to it and verifies it was received with valid v02 structure
3. Fails if the v02 export path crashes

suggestion

A minimal addition to the dynamic flow test (or a standalone test) that:

• Configures a post component with post_topicPrefix v02.post
• Posts at least one message
• Has a subscriber with topicPrefix v02.post that receives it
• Validates the message was actually received (not just that counts match)

Alternatively, a targeted check in flow_check.sh that greps the sender logs for v02 export errors.

context

The regression fix is in MetPX/sarracenia#1629. The unit test there covers v2wrapper.Message with both plain dicts and Message objects. But a flow-level test in sr_insects would have caught this before it hit production.

[petersilva]

@robjarawan hold on... what is said above is false.
There are posters that specify v02.post as the topic_prefix, and the topic_prefix is supposed to determine the format.
So the bug is that... don't need new tests... some existing tests are not working as intended and need repair.

idefix% grep topic_prefix */* | grep v02
report/tsarra_f20.conf:topic_prefix v02.post
sender/tsource2send_f50.conf:post_topic_prefix v02.post
subscribe/cp_f61.conf:topic_prefix v02.post
subscribe/u_sftp_f60.conf:topic_prefix v02.post
idefix% 

so the sender has a post_topicPrefix (the names are different because the config templates are for sarra v2, and the test includes translating the config to sr3's format) Normally the post_topicPrefix being v02.post should trigger the post_format being v02 as well. It used to.

[petersilva]

I did:

./flow_setup.sh config

then checked out random revisions, and did sr3 show sender/tsource2send_f50. I see the derived post_format is wrong at least as far back as 3.00.56, so it's not a new bug. just one that crept in a while ago.

The work-around is to specify post_format v02 in a config file explicitly... but we should probably fix this.

[petersilva]

the hoped for behaviour:

• format of posted messages is called post_format in the global options.
• format of posted messages should be called format in the publisher options.
• format of messages to be posted should be resolved when the publisher init is called.
• the basis for determining the format should be:
  • if the post_format is not specified (is not present, or None, or set to "default") then
    • if post_topicPrefix has a format as the first element, then set format to that.
    • if there is neither post_format, nor post_topicPrefix, then look at the format from the settings. (barring other information, output should match input.)
    • if there is no format specified, then default to v03. (also accomplished by default for format setting being v03) so no code necessary for this.

Sender decides format.

Asking this this question: I'm thinking that the format we send messages in should always be a function of the destination we are sending those messages to (and not the format they were received in.) so the post_format is not based on the format of the received message. So it makes sense to resolve it in configuration parsing, when resolving the publisher.