Notes for each dependency
The overview lists the dependencies Keploy records and replays on Linux and on macOS. This page says what to know about some of them. It describes Keploy 3.9.4, as its code is written.
How Keploy tells dependencies apart
- Keploy looks at the first bytes of each connection, not at its port. Your database or service can listen on any port.
- It tries the protocols it knows in a fixed order, and uses the first one that matches.
- A MySQL server speaks first, so Keploy recognizes MySQL from the server's greeting.
- A connection that matches no protocol is saved as raw bytes. It replays only when your tests send exactly the same bytes again.
Encrypted connections
When a connection starts a TLS handshake, Keploy answers it with a certificate from its own certificate authority (CA), then reads the protocol inside. This is how Keploy reads HTTPS calls, and any other connection that starts with TLS.
You don't install or trust that CA yourself. Each run sets it up, as HTTPS: do I need to trust Keploy's certificate? explains for Linux, macOS and Java.
PostgreSQL
At replay, Keploy answers the login itself:
- It accepts TLS when the server accepted it at record, and refuses it when the server refused it.
- It lets the client in without checking its password.
MySQL
At replay, Keploy answers the login from the recording:
- It doesn't compare the password the client sends.
- When the recording holds both a refused login and an accepted one for the same server, the replay lets the login in.
Redis
The Keploy CLI records and replays Redis on Linux. On macOS, run the tests in Docker, as Platforms shows.
A worker that waits on a blocking pop (BLPOP, BRPOP, BLMOVE,
BRPOPLPUSH, BLMPOP, BZPOPMIN, BZPOPMAX, BZMPOP, or XREAD and
XREADGROUP with BLOCK) is often still waiting when the recording ends. With
KEPLOY_PROVEN=1:
- That last pop, which the recording holds with no answer, stands for at most one pop at replay, in whichever test it comes.
- No test fails when the replay's worker doesn't send that pop.
- A pop that got an answer, a timeout included, is checked in the test that recorded it, like any other call.
Amazon SQS
On Linux, Keploy understands SQS calls. With
KEPLOY_PROVEN=1, a queue consumer's polls
(ReceiveMessage) replay this way:
- A poll that found no message answers at most one poll. No test fails because the consumer polled fewer times at replay, where every answer comes at once.
- A poll after every recorded one is answered as a poll that found no message.
- The messages of a recorded poll are delivered once in each test set. If the consumer polls again and that recording would answer it a second time, the replay answers that the poll found no message, as SQS does once a message is received and deleted.
- A poll that delivered messages, or that got an error, is checked in the test that recorded it, like any other call.
On macOS, Keploy records SQS calls as plain HTTPS calls, without SQS's own matching. A poll beyond the recorded ones then has no recording to answer it. Run SQS tests on Linux or in Docker, as Platforms shows.
See also
- Check what it supports: each dependency on Linux and macOS.
- Network isolation in replay: what a replay does with a call the recording doesn't have.
- Times, dates and ids that change: what a replay does with a value that is new on every run.
- Troubleshooting: symptoms, causes and fixes.