CI "Connection reset by peer" - Fix ECONNRESET on Runners
By Kaveh Alemi·Latchkey
An already-open connection was abruptly closed by the other end (or something in between) with a TCP RST. The transfer was underway and then cut off - a transient failure that usually succeeds on retry.
What this error means
A download or API call fails partway through with Connection reset by peer, ECONNRESET, or curl: (56) Recv failure. The connection had been established, so it is not a DNS or firewall problem - something dropped it mid-stream.
CI log
curl: (56) Recv failure: Connection reset by peer
# or
Error: read ECONNRESET
at TLSWrap.onStreamRead (node:internal/stream_base_commons)
Common causes
The remote or a proxy dropped the connection
An overloaded server, a load-balancer recycling backends, or a proxy idle-timeout sends a RST mid-transfer. None of it reflects a problem in your build.
Transient network instability
A brief network glitch between the runner and the remote can break an in-flight connection, surfacing as a reset.
How to fix it
Retry the transfer
A reset is transient; a clean retry typically completes the transfer.
Terminal
curl --retry 5 --retry-delay 2 --retry-all-errors -fSL "$URL" -o out
Reduce strain on the connection
Lower download concurrency so you are not hammering one endpoint.
Pull from a mirror or pull-through cache that is less loaded.
For package managers, set their built-in retry counts higher.
How to prevent it
Use bounded retry-with-backoff around network transfers.
Prefer mirrors/caches over a single hot upstream.
Tune client and package-manager retry settings for CI.
Frequently asked questions
What causes ""Connection reset by peer""?
An overloaded server, a load-balancer recycling backends, or a proxy idle-timeout sends a RST mid-transfer. None of it reflects a problem in your build.
How do I fix "Connection reset by peer"?
A reset is transient; a clean retry typically completes the transfer.
Can Latchkey fix this automatically?
Yes. Latchkey runs your GitHub Actions on managed runners that detect this failure, apply the fix, and retry the job automatically - self-healing is on by default.
