Fix: ignore invalid object counts to prevent infinite full-scan loops

[dgilm]

In lmd v2.2.0, change detection (hasChanged(ctx)) relies on countFromServer() to compare remote object counts against the local cache.

If countFromServer() fails (e.g. backend does not support Stats: queries or an error occurs), it returns -1. This was previously treated as a real count, leading lmd to incorrectly assume the dataset had changed. The result was repeated InitAllTables(ctx) reloads and excessive CPU usage.

Changes

• Added a guard to ignore negative count values returned from countFromServer().
• Change detection now only triggers when the count is valid (>= 0) and different from the local cache.

Impact

• Prevents infinite reload loops and CPU spikes when Stats: queries are unsupported or fail.
• Minimal and safe change: no effect when valid counts are returned.

Probably fixes: #144

[sni]

just curios, which core does not support stats queries?

[dgilm]

just curios, which core does not support stats queries?

Shinken’s Livestatus implementation is notoriously spotty with Stats: across some tables

This pull request is essentially an attempt to address #144
Although we are not entirely sure this is the root cause, it is worth giving it a try (as I said before, this change has no side-effects when valid counts are returned)

[sni]

Sure, could you apply the patch to the latest master?
Seems like your patch is based on version 1.1.1 from 2017.

[sni]

thanks, lets see if it helps...