* launch query and prefetch events earlier to improve latency
* add nicer query endpoint to perf tests
(cherry picked from commit 9851a1de91)
* add tracing for when a sender is starting to send and event.
* support use of multiple client channels; fix range of partitions; support pipelined return of query results
* use custom serialization to compress query results
* fix bug in test
* Add random scheduler
* add loadmonitor component
* add scheduling options & filehash tweaks
* keep partition prefix consistent with all the other tracing
* fix bug (failed to update estimated load when remote activities complete)
* Add aggresive scheduler mode
* fixed a typo
* basic load monitor
* checkpoint
* fix bug in ETW tracing
* change conditions for load monitor reporting
* use full tracing
* trace more information for OffloadCommandReceived
* change tracing of task messages to provide more consistent results
* several package updates, including DurableTask.Core 2.5.6
* update to account for tracing changes
* remove load monitor interval and add idle message
* add latency field to ActivityCompleted and RemoteActivityResultReceived
* added offload algorithms based on waiting time
* fix overflow in filehash
* fixes to loadmonitor logic, autoscaler, and temporary info sender
* fix missing file and update scale tracing
* change loadmonitor hosting so it moves less often
* implement overlay of pending commands on offload estimation
* improve precision of load estimation and distinguish between stationary and mobile.
* remove unnecessary left over line of code.
* replace push based algo with pull based
* fix so we don't pull more than mobile
* fix dumb mistake
* fix formula
* need more conservative default estimate for completion time, otherwise offload is too aggressive
* remove unnecessary constants.
* add tracing for RTT
* tweak parameters for RTT and smoothing
* batchworker tracing for senders
* use array instead of dictionary, fix concurrent modification exception
* revise sender for load monitor events to send only latest state, and to do rate limiting
* fix tracing
* turn off all loadmonitor activity when the parameter ActivityScheduler is not set LoadMonitor
* implemented "Static" setting for ActivityScheduler
* reduce severity of tracing in loadmonitor
* make filehash scale configurable
* update names, remove old algorithms
* improve terminology and implement solicitation
* fixes and simplifications
* minor updates
* simplify data structure for local activities
* use smarter concurrency defaults, same as for default backend
* fix tracing in LoadMonitor
* update filehash a bit to make it more uniform
* add test series for comparing local vs locavore
Co-authored-by: Sebastian Burckhardt <sburckha@microsoft.com>
* Go back to System.Threading.Channels version 4.7.1
* throw exception on 32bit process
* update to latest Microsoft.NET.Sdk.Functions
* use trace verbosity for unit tests
* revisit the mechanism for specifying alternate page blob storage: use more descriptive names, and implement deletion
* fix missing code.
* fix bug
* fix checkpoint devices so they use page blob storage as well.
(cherry picked from commit 1ea23e69899798e61f54eeba9dfeacbac2d4c483)
* do not trace batchworker work cycles where the batch is empty
* use 'details' instead of 'reason' for checkpoint events
* include partition id in eventhubs traces
* use LatencyMs instead of Latency for batch worker tracing.
* Do not emit warnings for ReceiverDisconnectedExceptions observed inside EventProcessorHost
* fix implementation of local file storage for FASTER
* fix test configuration in DurableTask.Netherite.AzureFunctions
* generalize the CI pipeline so it runs all tests again.
Sebastian Burckhardt
Christian Lechner
Xiangfeng Zhu