Reduce health check output (#10859)

This will make the health check clients 'silent'.
Use `IsNetworkOrHostDown` determine if network is ok so it mimics the functionality in the actual client.
This commit is contained in:
Klaus Post 2020-11-10 09:28:23 -08:00 committed by GitHub
parent cbdab62c1e
commit 06899210a7
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
4 changed files with 23 additions and 20 deletions

View File

@ -19,7 +19,6 @@ package cmd
import (
"bytes"
"context"
"errors"
"io"
"net/url"
"strconv"
@ -28,6 +27,7 @@ import (
xhttp "github.com/minio/minio/cmd/http"
"github.com/minio/minio/cmd/rest"
"github.com/minio/minio/pkg/dsync"
xnet "github.com/minio/minio/pkg/net"
)
// lockRESTClient is authenticable lock REST client
@ -153,15 +153,15 @@ func newlockRESTClient(endpoint Endpoint) *lockRESTClient {
restClient := rest.NewClient(serverURL, globalInternodeTransport, newAuthToken)
restClient.ExpectTimeouts = true
// Use a separate client to avoid recursive calls.
healthClient := rest.NewClient(serverURL, globalInternodeTransport, newAuthToken)
healthClient.ExpectTimeouts = true
restClient.HealthCheckFn = func() bool {
ctx, cancel := context.WithTimeout(GlobalContext, restClient.HealthCheckTimeout)
// Instantiate a new rest client for healthcheck
// to avoid recursive healthCheckFn()
respBody, err := rest.NewClient(serverURL, globalInternodeTransport, newAuthToken).Call(ctx, lockRESTMethodHealth, nil, nil, -1)
defer cancel()
respBody, err := healthClient.Call(ctx, lockRESTMethodHealth, nil, nil, -1)
xhttp.DrainBody(respBody)
cancel()
var ne *rest.NetworkError
return !errors.Is(err, context.DeadlineExceeded) && !errors.As(err, &ne)
return !xnet.IsNetworkOrHostDown(err, false)
}
return &lockRESTClient{endpoint: endpoint, restClient: restClient}

View File

@ -872,16 +872,17 @@ func newPeerRESTClient(peer *xnet.Host) *peerRESTClient {
}
restClient := rest.NewClient(serverURL, globalInternodeTransport, newAuthToken)
// Use a separate client to avoid recursive calls.
healthClient := rest.NewClient(serverURL, globalInternodeTransport, newAuthToken)
healthClient.ExpectTimeouts = true
// Construct a new health function.
restClient.HealthCheckFn = func() bool {
ctx, cancel := context.WithTimeout(GlobalContext, restClient.HealthCheckTimeout)
// Instantiate a new rest client for healthcheck
// to avoid recursive healthCheckFn()
respBody, err := rest.NewClient(serverURL, globalInternodeTransport, newAuthToken).Call(ctx, peerRESTMethodHealth, nil, nil, -1)
defer cancel()
respBody, err := healthClient.Call(ctx, peerRESTMethodHealth, nil, nil, -1)
xhttp.DrainBody(respBody)
cancel()
var ne *rest.NetworkError
return !errors.Is(err, context.DeadlineExceeded) && !errors.As(err, &ne)
return !xnet.IsNetworkOrHostDown(err, false)
}
return &peerRESTClient{host: peer, restClient: restClient}

View File

@ -117,7 +117,7 @@ func (c *Client) Call(ctx context.Context, method string, values url.Values, bod
}
resp, err := c.httpClient.Do(req)
if err != nil {
if xnet.IsNetworkOrHostDown(err, c.ExpectTimeouts) {
if c.HealthCheckFn != nil && xnet.IsNetworkOrHostDown(err, c.ExpectTimeouts) {
logger.Info("Marking %s temporary offline; caused by %v", c.url.String(), err)
c.MarkOffline()
}
@ -140,7 +140,7 @@ func (c *Client) Call(ctx context.Context, method string, values url.Values, bod
// with the caller to take the client offline purpose
// fully it should make sure to respond with '412'
// instead, see cmd/storage-rest-server.go for ideas.
if resp.StatusCode == http.StatusPreconditionFailed {
if c.HealthCheckFn != nil && resp.StatusCode == http.StatusPreconditionFailed {
logger.Info("Marking %s temporary offline; caused by PreconditionFailed.", c.url.String())
c.MarkOffline()
}
@ -148,7 +148,7 @@ func (c *Client) Call(ctx context.Context, method string, values url.Values, bod
// Limit the ReadAll(), just in case, because of a bug, the server responds with large data.
b, err := ioutil.ReadAll(io.LimitReader(resp.Body, c.MaxErrResponseSize))
if err != nil {
if xnet.IsNetworkOrHostDown(err, c.ExpectTimeouts) {
if c.HealthCheckFn != nil && xnet.IsNetworkOrHostDown(err, c.ExpectTimeouts) {
logger.Info("Marking %s temporary offline; caused by %v", c.url.String(), err)
c.MarkOffline()
}

View File

@ -615,15 +615,17 @@ func newStorageRESTClient(endpoint Endpoint, healthcheck bool) *storageRESTClien
}
restClient := rest.NewClient(serverURL, globalInternodeTransport, newAuthToken)
if healthcheck {
// Use a separate client to avoid recursive calls.
healthClient := rest.NewClient(serverURL, globalInternodeTransport, newAuthToken)
healthClient.ExpectTimeouts = true
restClient.HealthCheckFn = func() bool {
ctx, cancel := context.WithTimeout(GlobalContext, restClient.HealthCheckTimeout)
// Instantiate a new rest client for healthcheck
// to avoid recursive healthCheckFn()
respBody, err := rest.NewClient(serverURL, globalInternodeTransport, newAuthToken).Call(ctx, storageRESTMethodHealth, nil, nil, -1)
respBody, err := healthClient.Call(ctx, storageRESTMethodHealth, nil, nil, -1)
xhttp.DrainBody(respBody)
cancel()
return !errors.Is(err, context.DeadlineExceeded) && toStorageErr(err) != errDiskNotFound
return !xnet.IsNetworkOrHostDown(err, false) && toStorageErr(err) != errDiskNotFound
}
}