Be notified of new releases
Create your free GitHub account today to subscribe to this repository for new releases and build software alongside 31 million developers.Sign up
- TTL for servers in Rendezvous (#1773).
Previously in a client tries to resolve remote server in the Rendezvous and that server is absent - the client is hanged for some period of time. This was reported as inconvenient because if such server has never been published it is meaningless even to try to wait for it. Moreover, when a server is resolved it cleans its announce mapping until the next announce, which causes flapping in the info handler. Now all servers are considered live and will not be removed from the info handle if they are capable to reannounce themselves in ~30s. Otherwise the immediate "no peer found" error will be returned to the client.
- Memory leak in Optimus (#1785).
Reduced default branch-bound model thresholds. Switch for default optimization method while calculating current worker's price.
- Smart contracts migrations (#1778).
This PR is intended to provide more clear and stable migration process:
- Migration artifacts now are saved in migration_artifacts folder to maintain migration data.
- Migration contract from truffle suit is used.
- All current migrations are packed into one, as we start from the beginning.
- Useful makefile targets introduced.
- npm ci command is used for build stability.
- Properly detect for "red" cards in monitoring handler (#1776).
Now we have a situation when we have on-board and external video on the machine. Embedded graphics adapter also can be detected as DRI device, but does not provide any metrics interface. For GPU tuner plugin we're using a crutch: detecting device twice, via DRI and OpenCL, then comparing the results. This commit re-use that code to collect cards for monitoring interface.
- Drop nvidia socket if exists (#1772).
If a worker terminates unexpectedly, it leaves an orphaned socket file made by the NVidia GPU plugin. In such case the systemd cannot restart worker because of "address already in use" error. So we should remove this socket when initializing new listener for the GPU plugin.
- Blockchain Events api from block processing (#1771).
This PR fix behaviour which if fromBlock is not zero, events processing is unreachable, because "fromBlock" always greater to 1 than "toBlock".
- Proper RAM metrics collection (#1769).
- Close QUIC sessions and streams (#1770).
This fixes the memory leak in Rendezvous, Node and Worker components.
- Proper disk metrics (#1768).