Matteo Scandolo | a428586 | 2020-12-01 18:10:10 -0800 | [diff] [blame] | 1 | /* |
| 2 | Copyright 2014 The Kubernetes Authors. |
| 3 | |
| 4 | Licensed under the Apache License, Version 2.0 (the "License"); |
| 5 | you may not use this file except in compliance with the License. |
| 6 | You may obtain a copy of the License at |
| 7 | |
| 8 | http://www.apache.org/licenses/LICENSE-2.0 |
| 9 | |
| 10 | Unless required by applicable law or agreed to in writing, software |
| 11 | distributed under the License is distributed on an "AS IS" BASIS, |
| 12 | WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. |
| 13 | See the License for the specific language governing permissions and |
| 14 | limitations under the License. |
| 15 | */ |
| 16 | |
| 17 | package runtime |
| 18 | |
| 19 | import ( |
| 20 | "fmt" |
| 21 | "net/http" |
| 22 | "runtime" |
| 23 | "sync" |
| 24 | "time" |
| 25 | |
| 26 | "k8s.io/klog/v2" |
| 27 | ) |
| 28 | |
| 29 | var ( |
| 30 | // ReallyCrash controls the behavior of HandleCrash and now defaults |
| 31 | // true. It's still exposed so components can optionally set to false |
| 32 | // to restore prior behavior. |
| 33 | ReallyCrash = true |
| 34 | ) |
| 35 | |
| 36 | // PanicHandlers is a list of functions which will be invoked when a panic happens. |
| 37 | var PanicHandlers = []func(interface{}){logPanic} |
| 38 | |
| 39 | // HandleCrash simply catches a crash and logs an error. Meant to be called via |
| 40 | // defer. Additional context-specific handlers can be provided, and will be |
| 41 | // called in case of panic. HandleCrash actually crashes, after calling the |
| 42 | // handlers and logging the panic message. |
| 43 | // |
| 44 | // E.g., you can provide one or more additional handlers for something like shutting down go routines gracefully. |
| 45 | func HandleCrash(additionalHandlers ...func(interface{})) { |
| 46 | if r := recover(); r != nil { |
| 47 | for _, fn := range PanicHandlers { |
| 48 | fn(r) |
| 49 | } |
| 50 | for _, fn := range additionalHandlers { |
| 51 | fn(r) |
| 52 | } |
| 53 | if ReallyCrash { |
| 54 | // Actually proceed to panic. |
| 55 | panic(r) |
| 56 | } |
| 57 | } |
| 58 | } |
| 59 | |
| 60 | // logPanic logs the caller tree when a panic occurs (except in the special case of http.ErrAbortHandler). |
| 61 | func logPanic(r interface{}) { |
| 62 | if r == http.ErrAbortHandler { |
| 63 | // honor the http.ErrAbortHandler sentinel panic value: |
| 64 | // ErrAbortHandler is a sentinel panic value to abort a handler. |
| 65 | // While any panic from ServeHTTP aborts the response to the client, |
| 66 | // panicking with ErrAbortHandler also suppresses logging of a stack trace to the server's error log. |
| 67 | return |
| 68 | } |
| 69 | |
| 70 | // Same as stdlib http server code. Manually allocate stack trace buffer size |
| 71 | // to prevent excessively large logs |
| 72 | const size = 64 << 10 |
| 73 | stacktrace := make([]byte, size) |
| 74 | stacktrace = stacktrace[:runtime.Stack(stacktrace, false)] |
| 75 | if _, ok := r.(string); ok { |
| 76 | klog.Errorf("Observed a panic: %s\n%s", r, stacktrace) |
| 77 | } else { |
| 78 | klog.Errorf("Observed a panic: %#v (%v)\n%s", r, r, stacktrace) |
| 79 | } |
| 80 | } |
| 81 | |
| 82 | // ErrorHandlers is a list of functions which will be invoked when an unreturnable |
| 83 | // error occurs. |
| 84 | // TODO(lavalamp): for testability, this and the below HandleError function |
| 85 | // should be packaged up into a testable and reusable object. |
| 86 | var ErrorHandlers = []func(error){ |
| 87 | logError, |
| 88 | (&rudimentaryErrorBackoff{ |
| 89 | lastErrorTime: time.Now(), |
| 90 | // 1ms was the number folks were able to stomach as a global rate limit. |
| 91 | // If you need to log errors more than 1000 times a second you |
| 92 | // should probably consider fixing your code instead. :) |
| 93 | minPeriod: time.Millisecond, |
| 94 | }).OnError, |
| 95 | } |
| 96 | |
| 97 | // HandlerError is a method to invoke when a non-user facing piece of code cannot |
| 98 | // return an error and needs to indicate it has been ignored. Invoking this method |
| 99 | // is preferable to logging the error - the default behavior is to log but the |
| 100 | // errors may be sent to a remote server for analysis. |
| 101 | func HandleError(err error) { |
| 102 | // this is sometimes called with a nil error. We probably shouldn't fail and should do nothing instead |
| 103 | if err == nil { |
| 104 | return |
| 105 | } |
| 106 | |
| 107 | for _, fn := range ErrorHandlers { |
| 108 | fn(err) |
| 109 | } |
| 110 | } |
| 111 | |
| 112 | // logError prints an error with the call stack of the location it was reported |
| 113 | func logError(err error) { |
| 114 | klog.ErrorDepth(2, err) |
| 115 | } |
| 116 | |
| 117 | type rudimentaryErrorBackoff struct { |
| 118 | minPeriod time.Duration // immutable |
| 119 | // TODO(lavalamp): use the clock for testability. Need to move that |
| 120 | // package for that to be accessible here. |
| 121 | lastErrorTimeLock sync.Mutex |
| 122 | lastErrorTime time.Time |
| 123 | } |
| 124 | |
| 125 | // OnError will block if it is called more often than the embedded period time. |
| 126 | // This will prevent overly tight hot error loops. |
| 127 | func (r *rudimentaryErrorBackoff) OnError(error) { |
| 128 | r.lastErrorTimeLock.Lock() |
| 129 | defer r.lastErrorTimeLock.Unlock() |
| 130 | d := time.Since(r.lastErrorTime) |
| 131 | if d < r.minPeriod { |
| 132 | // If the time moves backwards for any reason, do nothing |
| 133 | time.Sleep(r.minPeriod - d) |
| 134 | } |
| 135 | r.lastErrorTime = time.Now() |
| 136 | } |
| 137 | |
| 138 | // GetCaller returns the caller of the function that calls it. |
| 139 | func GetCaller() string { |
| 140 | var pc [1]uintptr |
| 141 | runtime.Callers(3, pc[:]) |
| 142 | f := runtime.FuncForPC(pc[0]) |
| 143 | if f == nil { |
| 144 | return fmt.Sprintf("Unable to find caller") |
| 145 | } |
| 146 | return f.Name() |
| 147 | } |
| 148 | |
| 149 | // RecoverFromPanic replaces the specified error with an error containing the |
| 150 | // original error, and the call tree when a panic occurs. This enables error |
| 151 | // handlers to handle errors and panics the same way. |
| 152 | func RecoverFromPanic(err *error) { |
| 153 | if r := recover(); r != nil { |
| 154 | // Same as stdlib http server code. Manually allocate stack trace buffer size |
| 155 | // to prevent excessively large logs |
| 156 | const size = 64 << 10 |
| 157 | stacktrace := make([]byte, size) |
| 158 | stacktrace = stacktrace[:runtime.Stack(stacktrace, false)] |
| 159 | |
| 160 | *err = fmt.Errorf( |
| 161 | "recovered from panic %q. (err=%v) Call stack:\n%s", |
| 162 | r, |
| 163 | *err, |
| 164 | stacktrace) |
| 165 | } |
| 166 | } |
| 167 | |
| 168 | // Must panics on non-nil errors. Useful to handling programmer level errors. |
| 169 | func Must(err error) { |
| 170 | if err != nil { |
| 171 | panic(err) |
| 172 | } |
| 173 | } |