Description
When the SSE streaming connection drops with an EOFError, the SDK's error handling re-raises the exception as a RuntimeError, which propagates out and kills the streaming thread. This prevents the SyncManager from receiving a PUSH_RETRYABLE_ERROR notification, so the SDK never falls back to polling or attempts to reconnect.
In our production environment, we see ~25 EOFError events per day, each generating 4 log lines (including raw puts to stdout). The connections eventually recover on a timer, but the reconnection path through push_status_handler is bypassed.
SDK Version
Tested on 8.10.1, but the same pattern exists on master.
Root Cause
In lib/splitclient-rb/sse/event_source/client.rb, the connect_stream method has nested error handling for EOFError:
# Inner rescue (around line 109)
rescue EOFError => e
puts "SSE read operation EOF Exception!: #{e.inspect}"
@config.logger.error("SSE read operation EOF Exception!: #{e.inspect}")
raise 'eof exception' # re-raises as RuntimeError
# Outer rescue (around line 130)
rescue RuntimeError
raise 'eof exception' # propagates out, kills the threadThe double-raise means the exception escapes the thread entirely. The SyncManager's push_status_handler never receives PUSH_RETRYABLE_ERROR, so the reconnection/polling fallback path is never triggered.
Additionally, the puts on line 110 writes directly to stdout, bypassing the configured logger.
Expected Behavior
An EOFError during SSE streaming should:
- Log the event through the configured logger (not
puts)
- Notify the
SyncManager via PUSH_RETRYABLE_ERROR so it can trigger reconnection or fall back to polling
- Not kill the streaming thread
I see PR #622 touches some of the reconnection logic, but the specific double-raise pattern and puts call remain.
Suggested Fix
Replace the re-raise with a push_status notification, similar to how other connection errors are handled:
rescue EOFError => e
@config.logger.error("SSE EOF: #{e.inspect}")
@on[:action].call(Splits::EventSource::Actions::PUSH_RETRYABLE_ERROR)Happy to submit a PR for either or both of these if helpful.
Description
When the SSE streaming connection drops with an
EOFError, the SDK's error handling re-raises the exception as aRuntimeError, which propagates out and kills the streaming thread. This prevents theSyncManagerfrom receiving aPUSH_RETRYABLE_ERRORnotification, so the SDK never falls back to polling or attempts to reconnect.In our production environment, we see ~25 EOFError events per day, each generating 4 log lines (including raw
putsto stdout). The connections eventually recover on a timer, but the reconnection path throughpush_status_handleris bypassed.SDK Version
Tested on 8.10.1, but the same pattern exists on master.
Root Cause
In
lib/splitclient-rb/sse/event_source/client.rb, theconnect_streammethod has nested error handling forEOFError:The double-raise means the exception escapes the thread entirely. The
SyncManager'spush_status_handlernever receivesPUSH_RETRYABLE_ERROR, so the reconnection/polling fallback path is never triggered.Additionally, the
putson line 110 writes directly to stdout, bypassing the configured logger.Expected Behavior
An
EOFErrorduring SSE streaming should:puts)SyncManagerviaPUSH_RETRYABLE_ERRORso it can trigger reconnection or fall back to pollingI see PR #622 touches some of the reconnection logic, but the specific double-raise pattern and
putscall remain.Suggested Fix
Replace the re-raise with a
push_statusnotification, similar to how other connection errors are handled:Happy to submit a PR for either or both of these if helpful.